imsm: fix family number handling
[thirdparty/mdadm.git] / super-ddf.c
1 /*
2  * mdadm - manage Linux "md" devices aka RAID arrays.
3  *
4  * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
5  *
6  *
7  *    This program is free software; you can redistribute it and/or modify
8  *    it under the terms of the GNU General Public License as published by
9  *    the Free Software Foundation; either version 2 of the License, or
10  *    (at your option) any later version.
11  *
12  *    This program is distributed in the hope that it will be useful,
13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *    GNU General Public License for more details.
16  *
17  *    You should have received a copy of the GNU General Public License
18  *    along with this program; if not, write to the Free Software
19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
20  *
21  *    Author: Neil Brown
22  *    Email: <neil@brown.name>
23  *
24  * Specifications for DDF takes from Common RAID DDF Specification Revision 1.2
25  * (July 28 2006).  Reused by permission of SNIA.
26  */
27
28 #define HAVE_STDINT_H 1
29 #include "mdadm.h"
30 #include "mdmon.h"
31 #include "sha1.h"
32 #include <values.h>
33
34 /* a non-official T10 name for creation GUIDs */
35 static char T10[] = "Linux-MD";
36
37 /* DDF timestamps are 1980 based, so we need to add
38  * second-in-decade-of-seventies to convert to linux timestamps.
39  * 10 years with 2 leap years.
40  */
41 #define DECADE (3600*24*(365*10+2))
42 unsigned long crc32(
43         unsigned long crc,
44         const unsigned char *buf,
45         unsigned len);
46
47 /* The DDF metadata handling.
48  * DDF metadata lives at the end of the device.
49  * The last 512 byte block provides an 'anchor' which is used to locate
50  * the rest of the metadata which usually lives immediately behind the anchor.
51  *
52  * Note:
53  *  - all multibyte numeric fields are bigendian.
54  *  - all strings are space padded.
55  *
56  */
57
58 /* Primary Raid Level (PRL) */
59 #define DDF_RAID0       0x00
60 #define DDF_RAID1       0x01
61 #define DDF_RAID3       0x03
62 #define DDF_RAID4       0x04
63 #define DDF_RAID5       0x05
64 #define DDF_RAID1E      0x11
65 #define DDF_JBOD        0x0f
66 #define DDF_CONCAT      0x1f
67 #define DDF_RAID5E      0x15
68 #define DDF_RAID5EE     0x25
69 #define DDF_RAID6       0x06
70
71 /* Raid Level Qualifier (RLQ) */
72 #define DDF_RAID0_SIMPLE        0x00
73 #define DDF_RAID1_SIMPLE        0x00 /* just 2 devices in this plex */
74 #define DDF_RAID1_MULTI         0x01 /* exactly 3 devices in this plex */
75 #define DDF_RAID3_0             0x00 /* parity in first extent */
76 #define DDF_RAID3_N             0x01 /* parity in last extent */
77 #define DDF_RAID4_0             0x00 /* parity in first extent */
78 #define DDF_RAID4_N             0x01 /* parity in last extent */
79 /* these apply to raid5e and raid5ee as well */
80 #define DDF_RAID5_0_RESTART     0x00 /* same as 'right asymmetric' - layout 1 */
81 #define DDF_RAID6_0_RESTART     0x01 /* raid6 different from raid5 here!!! */
82 #define DDF_RAID5_N_RESTART     0x02 /* same as 'left asymmetric' - layout 0 */
83 #define DDF_RAID5_N_CONTINUE    0x03 /* same as 'left symmetric' - layout 2 */
84
85 #define DDF_RAID1E_ADJACENT     0x00 /* raid10 nearcopies==2 */
86 #define DDF_RAID1E_OFFSET       0x01 /* raid10 offsetcopies==2 */
87
88 /* Secondary RAID Level (SRL) */
89 #define DDF_2STRIPED    0x00    /* This is weirder than RAID0 !! */
90 #define DDF_2MIRRORED   0x01
91 #define DDF_2CONCAT     0x02
92 #define DDF_2SPANNED    0x03    /* This is also weird - be careful */
93
94 /* Magic numbers */
95 #define DDF_HEADER_MAGIC        __cpu_to_be32(0xDE11DE11)
96 #define DDF_CONTROLLER_MAGIC    __cpu_to_be32(0xAD111111)
97 #define DDF_PHYS_RECORDS_MAGIC  __cpu_to_be32(0x22222222)
98 #define DDF_PHYS_DATA_MAGIC     __cpu_to_be32(0x33333333)
99 #define DDF_VIRT_RECORDS_MAGIC  __cpu_to_be32(0xDDDDDDDD)
100 #define DDF_VD_CONF_MAGIC       __cpu_to_be32(0xEEEEEEEE)
101 #define DDF_SPARE_ASSIGN_MAGIC  __cpu_to_be32(0x55555555)
102 #define DDF_VU_CONF_MAGIC       __cpu_to_be32(0x88888888)
103 #define DDF_VENDOR_LOG_MAGIC    __cpu_to_be32(0x01dBEEF0)
104 #define DDF_BBM_LOG_MAGIC       __cpu_to_be32(0xABADB10C)
105
106 #define DDF_GUID_LEN    24
107 #define DDF_REVISION_0  "01.00.00"
108 #define DDF_REVISION_2  "01.02.00"
109
110 struct ddf_header {
111         __u32   magic;          /* DDF_HEADER_MAGIC */
112         __u32   crc;
113         char    guid[DDF_GUID_LEN];
114         char    revision[8];    /* 01.02.00 */
115         __u32   seq;            /* starts at '1' */
116         __u32   timestamp;
117         __u8    openflag;
118         __u8    foreignflag;
119         __u8    enforcegroups;
120         __u8    pad0;           /* 0xff */
121         __u8    pad1[12];       /* 12 * 0xff */
122         /* 64 bytes so far */
123         __u8    header_ext[32]; /* reserved: fill with 0xff */
124         __u64   primary_lba;
125         __u64   secondary_lba;
126         __u8    type;
127         __u8    pad2[3];        /* 0xff */
128         __u32   workspace_len;  /* sectors for vendor space -
129                                  * at least 32768(sectors) */
130         __u64   workspace_lba;
131         __u16   max_pd_entries; /* one of 15, 63, 255, 1023, 4095 */
132         __u16   max_vd_entries; /* 2^(4,6,8,10,12)-1 : i.e. as above */
133         __u16   max_partitions; /* i.e. max num of configuration
134                                    record entries per disk */
135         __u16   config_record_len; /* 1 +ROUNDUP(max_primary_element_entries
136                                                  *12/512) */
137         __u16   max_primary_element_entries; /* 16, 64, 256, 1024, or 4096 */
138         __u8    pad3[54];       /* 0xff */
139         /* 192 bytes so far */
140         __u32   controller_section_offset;
141         __u32   controller_section_length;
142         __u32   phys_section_offset;
143         __u32   phys_section_length;
144         __u32   virt_section_offset;
145         __u32   virt_section_length;
146         __u32   config_section_offset;
147         __u32   config_section_length;
148         __u32   data_section_offset;
149         __u32   data_section_length;
150         __u32   bbm_section_offset;
151         __u32   bbm_section_length;
152         __u32   diag_space_offset;
153         __u32   diag_space_length;
154         __u32   vendor_offset;
155         __u32   vendor_length;
156         /* 256 bytes so far */
157         __u8    pad4[256];      /* 0xff */
158 };
159
160 /* type field */
161 #define DDF_HEADER_ANCHOR       0x00
162 #define DDF_HEADER_PRIMARY      0x01
163 #define DDF_HEADER_SECONDARY    0x02
164
165 /* The content of the 'controller section' - global scope */
166 struct ddf_controller_data {
167         __u32   magic;                  /* DDF_CONTROLLER_MAGIC */
168         __u32   crc;
169         char    guid[DDF_GUID_LEN];
170         struct controller_type {
171                 __u16 vendor_id;
172                 __u16 device_id;
173                 __u16 sub_vendor_id;
174                 __u16 sub_device_id;
175         } type;
176         char    product_id[16];
177         __u8    pad[8]; /* 0xff */
178         __u8    vendor_data[448];
179 };
180
181 /* The content of phys_section - global scope */
182 struct phys_disk {
183         __u32   magic;          /* DDF_PHYS_RECORDS_MAGIC */
184         __u32   crc;
185         __u16   used_pdes;
186         __u16   max_pdes;
187         __u8    pad[52];
188         struct phys_disk_entry {
189                 char    guid[DDF_GUID_LEN];
190                 __u32   refnum;
191                 __u16   type;
192                 __u16   state;
193                 __u64   config_size; /* DDF structures must be after here */
194                 char    path[18];       /* another horrible structure really */
195                 __u8    pad[6];
196         } entries[0];
197 };
198
199 /* phys_disk_entry.type is a bitmap - bigendian remember */
200 #define DDF_Forced_PD_GUID              1
201 #define DDF_Active_in_VD                2
202 #define DDF_Global_Spare                4 /* VD_CONF records are ignored */
203 #define DDF_Spare                       8 /* overrides Global_spare */
204 #define DDF_Foreign                     16
205 #define DDF_Legacy                      32 /* no DDF on this device */
206
207 #define DDF_Interface_mask              0xf00
208 #define DDF_Interface_SCSI              0x100
209 #define DDF_Interface_SAS               0x200
210 #define DDF_Interface_SATA              0x300
211 #define DDF_Interface_FC                0x400
212
213 /* phys_disk_entry.state is a bigendian bitmap */
214 #define DDF_Online                      1
215 #define DDF_Failed                      2 /* overrides  1,4,8 */
216 #define DDF_Rebuilding                  4
217 #define DDF_Transition                  8
218 #define DDF_SMART                       16
219 #define DDF_ReadErrors                  32
220 #define DDF_Missing                     64
221
222 /* The content of the virt_section global scope */
223 struct virtual_disk {
224         __u32   magic;          /* DDF_VIRT_RECORDS_MAGIC */
225         __u32   crc;
226         __u16   populated_vdes;
227         __u16   max_vdes;
228         __u8    pad[52];
229         struct virtual_entry {
230                 char    guid[DDF_GUID_LEN];
231                 __u16   unit;
232                 __u16   pad0;   /* 0xffff */
233                 __u16   guid_crc;
234                 __u16   type;
235                 __u8    state;
236                 __u8    init_state;
237                 __u8    pad1[14];
238                 char    name[16];
239         } entries[0];
240 };
241
242 /* virtual_entry.type is a bitmap - bigendian */
243 #define DDF_Shared              1
244 #define DDF_Enforce_Groups      2
245 #define DDF_Unicode             4
246 #define DDF_Owner_Valid         8
247
248 /* virtual_entry.state is a bigendian bitmap */
249 #define DDF_state_mask          0x7
250 #define DDF_state_optimal       0x0
251 #define DDF_state_degraded      0x1
252 #define DDF_state_deleted       0x2
253 #define DDF_state_missing       0x3
254 #define DDF_state_failed        0x4
255 #define DDF_state_part_optimal  0x5
256
257 #define DDF_state_morphing      0x8
258 #define DDF_state_inconsistent  0x10
259
260 /* virtual_entry.init_state is a bigendian bitmap */
261 #define DDF_initstate_mask      0x03
262 #define DDF_init_not            0x00
263 #define DDF_init_quick          0x01 /* initialisation is progress.
264                                       * i.e. 'state_inconsistent' */
265 #define DDF_init_full           0x02
266
267 #define DDF_access_mask         0xc0
268 #define DDF_access_rw           0x00
269 #define DDF_access_ro           0x80
270 #define DDF_access_blocked      0xc0
271
272 /* The content of the config_section - local scope
273  * It has multiple records each config_record_len sectors
274  * They can be vd_config or spare_assign
275  */
276
277 struct vd_config {
278         __u32   magic;          /* DDF_VD_CONF_MAGIC */
279         __u32   crc;
280         char    guid[DDF_GUID_LEN];
281         __u32   timestamp;
282         __u32   seqnum;
283         __u8    pad0[24];
284         __u16   prim_elmnt_count;
285         __u8    chunk_shift;    /* 0 == 512, 1==1024 etc */
286         __u8    prl;
287         __u8    rlq;
288         __u8    sec_elmnt_count;
289         __u8    sec_elmnt_seq;
290         __u8    srl;
291         __u64   blocks;         /* blocks per component could be different
292                                  * on different component devices...(only
293                                  * for concat I hope) */
294         __u64   array_blocks;   /* blocks in array */
295         __u8    pad1[8];
296         __u32   spare_refs[8];
297         __u8    cache_pol[8];
298         __u8    bg_rate;
299         __u8    pad2[3];
300         __u8    pad3[52];
301         __u8    pad4[192];
302         __u8    v0[32]; /* reserved- 0xff */
303         __u8    v1[32]; /* reserved- 0xff */
304         __u8    v2[16]; /* reserved- 0xff */
305         __u8    v3[16]; /* reserved- 0xff */
306         __u8    vendor[32];
307         __u32   phys_refnum[0]; /* refnum of each disk in sequence */
308       /*__u64   lba_offset[0];  LBA offset in each phys.  Note extents in a
309                                 bvd are always the same size */
310 };
311
312 /* vd_config.cache_pol[7] is a bitmap */
313 #define DDF_cache_writeback     1       /* else writethrough */
314 #define DDF_cache_wadaptive     2       /* only applies if writeback */
315 #define DDF_cache_readahead     4
316 #define DDF_cache_radaptive     8       /* only if doing read-ahead */
317 #define DDF_cache_ifnobatt      16      /* even to write cache if battery is poor */
318 #define DDF_cache_wallowed      32      /* enable write caching */
319 #define DDF_cache_rallowed      64      /* enable read caching */
320
321 struct spare_assign {
322         __u32   magic;          /* DDF_SPARE_ASSIGN_MAGIC */
323         __u32   crc;
324         __u32   timestamp;
325         __u8    reserved[7];
326         __u8    type;
327         __u16   populated;      /* SAEs used */
328         __u16   max;            /* max SAEs */
329         __u8    pad[8];
330         struct spare_assign_entry {
331                 char    guid[DDF_GUID_LEN];
332                 __u16   secondary_element;
333                 __u8    pad[6];
334         } spare_ents[0];
335 };
336 /* spare_assign.type is a bitmap */
337 #define DDF_spare_dedicated     0x1     /* else global */
338 #define DDF_spare_revertible    0x2     /* else committable */
339 #define DDF_spare_active        0x4     /* else not active */
340 #define DDF_spare_affinity      0x8     /* enclosure affinity */
341
342 /* The data_section contents - local scope */
343 struct disk_data {
344         __u32   magic;          /* DDF_PHYS_DATA_MAGIC */
345         __u32   crc;
346         char    guid[DDF_GUID_LEN];
347         __u32   refnum;         /* crc of some magic drive data ... */
348         __u8    forced_ref;     /* set when above was not result of magic */
349         __u8    forced_guid;    /* set if guid was forced rather than magic */
350         __u8    vendor[32];
351         __u8    pad[442];
352 };
353
354 /* bbm_section content */
355 struct bad_block_log {
356         __u32   magic;
357         __u32   crc;
358         __u16   entry_count;
359         __u32   spare_count;
360         __u8    pad[10];
361         __u64   first_spare;
362         struct mapped_block {
363                 __u64   defective_start;
364                 __u32   replacement_start;
365                 __u16   remap_count;
366                 __u8    pad[2];
367         } entries[0];
368 };
369
370 /* Struct for internally holding ddf structures */
371 /* The DDF structure stored on each device is potentially
372  * quite different, as some data is global and some is local.
373  * The global data is:
374  *   - ddf header
375  *   - controller_data
376  *   - Physical disk records
377  *   - Virtual disk records
378  * The local data is:
379  *   - Configuration records
380  *   - Physical Disk data section
381  *  (  and Bad block and vendor which I don't care about yet).
382  *
383  * The local data is parsed into separate lists as it is read
384  * and reconstructed for writing.  This means that we only need
385  * to make config changes once and they are automatically
386  * propagated to all devices.
387  * Note that the ddf_super has space of the conf and disk data
388  * for this disk and also for a list of all such data.
389  * The list is only used for the superblock that is being
390  * built in Create or Assemble to describe the whole array.
391  */
392 struct ddf_super {
393         struct ddf_header anchor, primary, secondary;
394         struct ddf_controller_data controller;
395         struct ddf_header *active;
396         struct phys_disk        *phys;
397         struct virtual_disk     *virt;
398         int pdsize, vdsize;
399         int max_part, mppe, conf_rec_len;
400         int currentdev;
401         int updates_pending;
402         struct vcl {
403                 union {
404                         char space[512];
405                         struct {
406                                 struct vcl      *next;
407                                 __u64           *lba_offset; /* location in 'conf' of
408                                                               * the lba table */
409                                 int     vcnum; /* index into ->virt */
410                                 __u64           *block_sizes; /* NULL if all the same */
411                         };
412                 };
413                 struct vd_config conf;
414         } *conflist, *currentconf;
415         struct dl {
416                 union {
417                         char space[512];
418                         struct {
419                                 struct dl       *next;
420                                 int major, minor;
421                                 char *devname;
422                                 int fd;
423                                 unsigned long long size; /* sectors */
424                                 int pdnum;      /* index in ->phys */
425                                 struct spare_assign *spare;
426                                 void *mdupdate; /* hold metadata update */
427
428                                 /* These fields used by auto-layout */
429                                 int raiddisk; /* slot to fill in autolayout */
430                                 __u64 esize;
431                         };
432                 };
433                 struct disk_data disk;
434                 struct vcl *vlist[0]; /* max_part in size */
435         } *dlist, *add_list;
436 };
437
438 #ifndef offsetof
439 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
440 #endif
441
442
443 static int calc_crc(void *buf, int len)
444 {
445         /* crcs are always at the same place as in the ddf_header */
446         struct ddf_header *ddf = buf;
447         __u32 oldcrc = ddf->crc;
448         __u32 newcrc;
449         ddf->crc = 0xffffffff;
450
451         newcrc = crc32(0, buf, len);
452         ddf->crc = oldcrc;
453         /* The crc is store (like everything) bigendian, so convert
454          * here for simplicity
455          */
456         return __cpu_to_be32(newcrc);
457 }
458
459 static int load_ddf_header(int fd, unsigned long long lba,
460                            unsigned long long size,
461                            int type,
462                            struct ddf_header *hdr, struct ddf_header *anchor)
463 {
464         /* read a ddf header (primary or secondary) from fd/lba
465          * and check that it is consistent with anchor
466          * Need to check:
467          *   magic, crc, guid, rev, and LBA's header_type, and
468          *  everything after header_type must be the same
469          */
470         if (lba >= size-1)
471                 return 0;
472
473         if (lseek64(fd, lba<<9, 0) < 0)
474                 return 0;
475
476         if (read(fd, hdr, 512) != 512)
477                 return 0;
478
479         if (hdr->magic != DDF_HEADER_MAGIC)
480                 return 0;
481         if (calc_crc(hdr, 512) != hdr->crc)
482                 return 0;
483         if (memcmp(anchor->guid, hdr->guid, DDF_GUID_LEN) != 0 ||
484             memcmp(anchor->revision, hdr->revision, 8) != 0 ||
485             anchor->primary_lba != hdr->primary_lba ||
486             anchor->secondary_lba != hdr->secondary_lba ||
487             hdr->type != type ||
488             memcmp(anchor->pad2, hdr->pad2, 512 -
489                    offsetof(struct ddf_header, pad2)) != 0)
490                 return 0;
491
492         /* Looks good enough to me... */
493         return 1;
494 }
495
496 static void *load_section(int fd, struct ddf_super *super, void *buf,
497                           __u32 offset_be, __u32 len_be, int check)
498 {
499         unsigned long long offset = __be32_to_cpu(offset_be);
500         unsigned long long len = __be32_to_cpu(len_be);
501         int dofree = (buf == NULL);
502
503         if (check)
504                 if (len != 2 && len != 8 && len != 32
505                     && len != 128 && len != 512)
506                         return NULL;
507
508         if (len > 1024)
509                 return NULL;
510         if (buf) {
511                 /* All pre-allocated sections are a single block */
512                 if (len != 1)
513                         return NULL;
514         } else if (posix_memalign(&buf, 512, len<<9) != 0)
515                 buf = NULL;
516
517         if (!buf)
518                 return NULL;
519
520         if (super->active->type == 1)
521                 offset += __be64_to_cpu(super->active->primary_lba);
522         else
523                 offset += __be64_to_cpu(super->active->secondary_lba);
524
525         if (lseek64(fd, offset<<9, 0) != (offset<<9)) {
526                 if (dofree)
527                         free(buf);
528                 return NULL;
529         }
530         if (read(fd, buf, len<<9) != (len<<9)) {
531                 if (dofree)
532                         free(buf);
533                 return NULL;
534         }
535         return buf;
536 }
537
538 static int load_ddf_headers(int fd, struct ddf_super *super, char *devname)
539 {
540         unsigned long long dsize;
541
542         get_dev_size(fd, NULL, &dsize);
543
544         if (lseek64(fd, dsize-512, 0) < 0) {
545                 if (devname)
546                         fprintf(stderr,
547                                 Name": Cannot seek to anchor block on %s: %s\n",
548                                 devname, strerror(errno));
549                 return 1;
550         }
551         if (read(fd, &super->anchor, 512) != 512) {
552                 if (devname)
553                         fprintf(stderr,
554                                 Name ": Cannot read anchor block on %s: %s\n",
555                                 devname, strerror(errno));
556                 return 1;
557         }
558         if (super->anchor.magic != DDF_HEADER_MAGIC) {
559                 if (devname)
560                         fprintf(stderr, Name ": no DDF anchor found on %s\n",
561                                 devname);
562                 return 2;
563         }
564         if (calc_crc(&super->anchor, 512) != super->anchor.crc) {
565                 if (devname)
566                         fprintf(stderr, Name ": bad CRC on anchor on %s\n",
567                                 devname);
568                 return 2;
569         }
570         if (memcmp(super->anchor.revision, DDF_REVISION_0, 8) != 0 &&
571             memcmp(super->anchor.revision, DDF_REVISION_2, 8) != 0) {
572                 if (devname)
573                         fprintf(stderr, Name ": can only support super revision"
574                                 " %.8s and earlier, not %.8s on %s\n",
575                                 DDF_REVISION_2, super->anchor.revision,devname);
576                 return 2;
577         }
578         if (load_ddf_header(fd, __be64_to_cpu(super->anchor.primary_lba),
579                             dsize >> 9,  1,
580                             &super->primary, &super->anchor) == 0) {
581                 if (devname)
582                         fprintf(stderr,
583                                 Name ": Failed to load primary DDF header "
584                                 "on %s\n", devname);
585                 return 2;
586         }
587         super->active = &super->primary;
588         if (load_ddf_header(fd, __be64_to_cpu(super->anchor.secondary_lba),
589                             dsize >> 9,  2,
590                             &super->secondary, &super->anchor)) {
591                 if ((__be32_to_cpu(super->primary.seq)
592                      < __be32_to_cpu(super->secondary.seq) &&
593                      !super->secondary.openflag)
594                     || (__be32_to_cpu(super->primary.seq)
595                         == __be32_to_cpu(super->secondary.seq) &&
596                         super->primary.openflag && !super->secondary.openflag)
597                         )
598                         super->active = &super->secondary;
599         }
600         return 0;
601 }
602
603 static int load_ddf_global(int fd, struct ddf_super *super, char *devname)
604 {
605         void *ok;
606         ok = load_section(fd, super, &super->controller,
607                           super->active->controller_section_offset,
608                           super->active->controller_section_length,
609                           0);
610         super->phys = load_section(fd, super, NULL,
611                                    super->active->phys_section_offset,
612                                    super->active->phys_section_length,
613                                    1);
614         super->pdsize = __be32_to_cpu(super->active->phys_section_length) * 512;
615
616         super->virt = load_section(fd, super, NULL,
617                                    super->active->virt_section_offset,
618                                    super->active->virt_section_length,
619                                    1);
620         super->vdsize = __be32_to_cpu(super->active->virt_section_length) * 512;
621         if (!ok ||
622             !super->phys ||
623             !super->virt) {
624                 free(super->phys);
625                 free(super->virt);
626                 super->phys = NULL;
627                 super->virt = NULL;
628                 return 2;
629         }
630         super->conflist = NULL;
631         super->dlist = NULL;
632
633         super->max_part = __be16_to_cpu(super->active->max_partitions);
634         super->mppe = __be16_to_cpu(super->active->max_primary_element_entries);
635         super->conf_rec_len = __be16_to_cpu(super->active->config_record_len);
636         return 0;
637 }
638
639 static int load_ddf_local(int fd, struct ddf_super *super,
640                           char *devname, int keep)
641 {
642         struct dl *dl;
643         struct stat stb;
644         char *conf;
645         int i;
646         int confsec;
647         int vnum;
648         int max_virt_disks = __be16_to_cpu(super->active->max_vd_entries);
649         unsigned long long dsize;
650
651         /* First the local disk info */
652         if (posix_memalign((void**)&dl, 512,
653                        sizeof(*dl) +
654                        (super->max_part) * sizeof(dl->vlist[0])) != 0) {
655                 fprintf(stderr, Name ": %s could not allocate disk info buffer\n",
656                         __func__);
657                 return 1;
658         }
659
660         load_section(fd, super, &dl->disk,
661                      super->active->data_section_offset,
662                      super->active->data_section_length,
663                      0);
664         dl->devname = devname ? strdup(devname) : NULL;
665
666         fstat(fd, &stb);
667         dl->major = major(stb.st_rdev);
668         dl->minor = minor(stb.st_rdev);
669         dl->next = super->dlist;
670         dl->fd = keep ? fd : -1;
671
672         dl->size = 0;
673         if (get_dev_size(fd, devname, &dsize))
674                 dl->size = dsize >> 9;
675         dl->spare = NULL;
676         for (i=0 ; i < super->max_part ; i++)
677                 dl->vlist[i] = NULL;
678         super->dlist = dl;
679         dl->pdnum = -1;
680         for (i=0; i < __be16_to_cpu(super->active->max_pd_entries); i++)
681                 if (memcmp(super->phys->entries[i].guid,
682                            dl->disk.guid, DDF_GUID_LEN) == 0)
683                         dl->pdnum = i;
684
685         /* Now the config list. */
686         /* 'conf' is an array of config entries, some of which are
687          * probably invalid.  Those which are good need to be copied into
688          * the conflist
689          */
690
691         conf = load_section(fd, super, NULL,
692                             super->active->config_section_offset,
693                             super->active->config_section_length,
694                             0);
695
696         vnum = 0;
697         for (confsec = 0;
698              confsec < __be32_to_cpu(super->active->config_section_length);
699              confsec += super->conf_rec_len) {
700                 struct vd_config *vd =
701                         (struct vd_config *)((char*)conf + confsec*512);
702                 struct vcl *vcl;
703
704                 if (vd->magic == DDF_SPARE_ASSIGN_MAGIC) {
705                         if (dl->spare)
706                                 continue;
707                         if (posix_memalign((void**)&dl->spare, 512,
708                                        super->conf_rec_len*512) != 0) {
709                                 fprintf(stderr, Name
710                                         ": %s could not allocate spare info buf\n",
711                                         __func__);
712                                 return 1;
713                         }
714                                 
715                         memcpy(dl->spare, vd, super->conf_rec_len*512);
716                         continue;
717                 }
718                 if (vd->magic != DDF_VD_CONF_MAGIC)
719                         continue;
720                 for (vcl = super->conflist; vcl; vcl = vcl->next) {
721                         if (memcmp(vcl->conf.guid,
722                                    vd->guid, DDF_GUID_LEN) == 0)
723                                 break;
724                 }
725
726                 if (vcl) {
727                         dl->vlist[vnum++] = vcl;
728                         if (__be32_to_cpu(vd->seqnum) <=
729                             __be32_to_cpu(vcl->conf.seqnum))
730                                 continue;
731                 } else {
732                         if (posix_memalign((void**)&vcl, 512,
733                                        (super->conf_rec_len*512 +
734                                         offsetof(struct vcl, conf))) != 0) {
735                                 fprintf(stderr, Name
736                                         ": %s could not allocate vcl buf\n",
737                                         __func__);
738                                 return 1;
739                         }
740                         vcl->next = super->conflist;
741                         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
742                         super->conflist = vcl;
743                         dl->vlist[vnum++] = vcl;
744                 }
745                 memcpy(&vcl->conf, vd, super->conf_rec_len*512);
746                 vcl->lba_offset = (__u64*)
747                         &vcl->conf.phys_refnum[super->mppe];
748
749                 for (i=0; i < max_virt_disks ; i++)
750                         if (memcmp(super->virt->entries[i].guid,
751                                    vcl->conf.guid, DDF_GUID_LEN)==0)
752                                 break;
753                 if (i < max_virt_disks)
754                         vcl->vcnum = i;
755         }
756         free(conf);
757
758         return 0;
759 }
760
761 #ifndef MDASSEMBLE
762 static int load_super_ddf_all(struct supertype *st, int fd,
763                               void **sbp, char *devname, int keep_fd);
764 #endif
765
766 static void free_super_ddf(struct supertype *st);
767
768 static int load_super_ddf(struct supertype *st, int fd,
769                           char *devname)
770 {
771         unsigned long long dsize;
772         struct ddf_super *super;
773         int rv;
774
775 #ifndef MDASSEMBLE
776         /* if 'fd' is a container, load metadata from all the devices */
777         if (load_super_ddf_all(st, fd, &st->sb, devname, 1) == 0)
778                 return 0;
779 #endif
780         if (st->subarray[0])
781                 return 1; /* FIXME Is this correct */
782
783         if (get_dev_size(fd, devname, &dsize) == 0)
784                 return 1;
785
786         /* 32M is a lower bound */
787         if (dsize <= 32*1024*1024) {
788                 if (devname)
789                         fprintf(stderr,
790                                 Name ": %s is too small for ddf: "
791                                 "size is %llu sectors.\n",
792                                 devname, dsize>>9);
793                 return 1;
794         }
795         if (dsize & 511) {
796                 if (devname)
797                         fprintf(stderr,
798                                 Name ": %s is an odd size for ddf: "
799                                 "size is %llu bytes.\n",
800                                 devname, dsize);
801                 return 1;
802         }
803
804         free_super_ddf(st);
805
806         if (posix_memalign((void**)&super, 512, sizeof(*super))!= 0) {
807                 fprintf(stderr, Name ": malloc of %zu failed.\n",
808                         sizeof(*super));
809                 return 1;
810         }
811         memset(super, 0, sizeof(*super));
812
813         rv = load_ddf_headers(fd, super, devname);
814         if (rv) {
815                 free(super);
816                 return rv;
817         }
818
819         /* Have valid headers and have chosen the best. Let's read in the rest*/
820
821         rv = load_ddf_global(fd, super, devname);
822
823         if (rv) {
824                 if (devname)
825                         fprintf(stderr,
826                                 Name ": Failed to load all information "
827                                 "sections on %s\n", devname);
828                 free(super);
829                 return rv;
830         }
831
832         rv = load_ddf_local(fd, super, devname, 0);
833
834         if (rv) {
835                 if (devname)
836                         fprintf(stderr,
837                                 Name ": Failed to load all information "
838                                 "sections on %s\n", devname);
839                 free(super);
840                 return rv;
841         }
842
843         if (st->subarray[0]) {
844                 struct vcl *v;
845
846                 for (v = super->conflist; v; v = v->next)
847                         if (v->vcnum == atoi(st->subarray))
848                                 super->currentconf = v;
849                 if (!super->currentconf) {
850                         free(super);
851                         return 1;
852                 }
853         }
854
855         /* Should possibly check the sections .... */
856
857         st->sb = super;
858         if (st->ss == NULL) {
859                 st->ss = &super_ddf;
860                 st->minor_version = 0;
861                 st->max_devs = 512;
862         }
863         st->loaded_container = 0;
864         return 0;
865
866 }
867
868 static void free_super_ddf(struct supertype *st)
869 {
870         struct ddf_super *ddf = st->sb;
871         if (ddf == NULL)
872                 return;
873         free(ddf->phys);
874         free(ddf->virt);
875         while (ddf->conflist) {
876                 struct vcl *v = ddf->conflist;
877                 ddf->conflist = v->next;
878                 if (v->block_sizes)
879                         free(v->block_sizes);
880                 free(v);
881         }
882         while (ddf->dlist) {
883                 struct dl *d = ddf->dlist;
884                 ddf->dlist = d->next;
885                 if (d->fd >= 0)
886                         close(d->fd);
887                 if (d->spare)
888                         free(d->spare);
889                 free(d);
890         }
891         free(ddf);
892         st->sb = NULL;
893 }
894
895 static struct supertype *match_metadata_desc_ddf(char *arg)
896 {
897         /* 'ddf' only support containers */
898         struct supertype *st;
899         if (strcmp(arg, "ddf") != 0 &&
900             strcmp(arg, "default") != 0
901                 )
902                 return NULL;
903
904         st = malloc(sizeof(*st));
905         memset(st, 0, sizeof(*st));
906         st->ss = &super_ddf;
907         st->max_devs = 512;
908         st->minor_version = 0;
909         st->sb = NULL;
910         return st;
911 }
912
913
914 #ifndef MDASSEMBLE
915
916 static mapping_t ddf_state[] = {
917         { "Optimal", 0},
918         { "Degraded", 1},
919         { "Deleted", 2},
920         { "Missing", 3},
921         { "Failed", 4},
922         { "Partially Optimal", 5},
923         { "-reserved-", 6},
924         { "-reserved-", 7},
925         { NULL, 0}
926 };
927
928 static mapping_t ddf_init_state[] = {
929         { "Not Initialised", 0},
930         { "QuickInit in Progress", 1},
931         { "Fully Initialised", 2},
932         { "*UNKNOWN*", 3},
933         { NULL, 0}
934 };
935 static mapping_t ddf_access[] = {
936         { "Read/Write", 0},
937         { "Reserved", 1},
938         { "Read Only", 2},
939         { "Blocked (no access)", 3},
940         { NULL ,0}
941 };
942
943 static mapping_t ddf_level[] = {
944         { "RAID0", DDF_RAID0},
945         { "RAID1", DDF_RAID1},
946         { "RAID3", DDF_RAID3},
947         { "RAID4", DDF_RAID4},
948         { "RAID5", DDF_RAID5},
949         { "RAID1E",DDF_RAID1E},
950         { "JBOD",  DDF_JBOD},
951         { "CONCAT",DDF_CONCAT},
952         { "RAID5E",DDF_RAID5E},
953         { "RAID5EE",DDF_RAID5EE},
954         { "RAID6", DDF_RAID6},
955         { NULL, 0}
956 };
957 static mapping_t ddf_sec_level[] = {
958         { "Striped", DDF_2STRIPED},
959         { "Mirrored", DDF_2MIRRORED},
960         { "Concat", DDF_2CONCAT},
961         { "Spanned", DDF_2SPANNED},
962         { NULL, 0}
963 };
964 #endif
965
966 struct num_mapping {
967         int num1, num2;
968 };
969 static struct num_mapping ddf_level_num[] = {
970         { DDF_RAID0, 0 },
971         { DDF_RAID1, 1 },
972         { DDF_RAID3, LEVEL_UNSUPPORTED },
973         { DDF_RAID4, 4 },
974         { DDF_RAID5, 5 },
975         { DDF_RAID1E, LEVEL_UNSUPPORTED },
976         { DDF_JBOD, LEVEL_UNSUPPORTED },
977         { DDF_CONCAT, LEVEL_LINEAR },
978         { DDF_RAID5E, LEVEL_UNSUPPORTED },
979         { DDF_RAID5EE, LEVEL_UNSUPPORTED },
980         { DDF_RAID6, 6},
981         { MAXINT, MAXINT }
982 };
983
984 static int map_num1(struct num_mapping *map, int num)
985 {
986         int i;
987         for (i=0 ; map[i].num1 != MAXINT; i++)
988                 if (map[i].num1 == num)
989                         break;
990         return map[i].num2;
991 }
992
993 static int all_ff(char *guid)
994 {
995         int i;
996         for (i = 0; i < DDF_GUID_LEN; i++)
997                 if (guid[i] != (char)0xff)
998                         return 0;
999         return 1;
1000 }
1001
1002 #ifndef MDASSEMBLE
1003 static void print_guid(char *guid, int tstamp)
1004 {
1005         /* A GUIDs are part (or all) ASCII and part binary.
1006          * They tend to be space padded.
1007          * We print the GUID in HEX, then in parentheses add
1008          * any initial ASCII sequence, and a possible
1009          * time stamp from bytes 16-19
1010          */
1011         int l = DDF_GUID_LEN;
1012         int i;
1013
1014         for (i=0 ; i<DDF_GUID_LEN ; i++) {
1015                 if ((i&3)==0 && i != 0) printf(":");
1016                 printf("%02X", guid[i]&255);
1017         }
1018
1019         printf("\n                  (");
1020         while (l && guid[l-1] == ' ')
1021                 l--;
1022         for (i=0 ; i<l ; i++) {
1023                 if (guid[i] >= 0x20 && guid[i] < 0x7f)
1024                         fputc(guid[i], stdout);
1025                 else
1026                         break;
1027         }
1028         if (tstamp) {
1029                 time_t then = __be32_to_cpu(*(__u32*)(guid+16)) + DECADE;
1030                 char tbuf[100];
1031                 struct tm *tm;
1032                 tm = localtime(&then);
1033                 strftime(tbuf, 100, " %D %T",tm);
1034                 fputs(tbuf, stdout);
1035         }
1036         printf(")");
1037 }
1038
1039 static void examine_vd(int n, struct ddf_super *sb, char *guid)
1040 {
1041         int crl = sb->conf_rec_len;
1042         struct vcl *vcl;
1043
1044         for (vcl = sb->conflist ; vcl ; vcl = vcl->next) {
1045                 int i;
1046                 struct vd_config *vc = &vcl->conf;
1047
1048                 if (calc_crc(vc, crl*512) != vc->crc)
1049                         continue;
1050                 if (memcmp(vc->guid, guid, DDF_GUID_LEN) != 0)
1051                         continue;
1052
1053                 /* Ok, we know about this VD, let's give more details */
1054                 printf(" Raid Devices[%d] : %d (", n,
1055                        __be16_to_cpu(vc->prim_elmnt_count));
1056                 for (i=0; i<__be16_to_cpu(vc->prim_elmnt_count); i++) {
1057                         int j;
1058                         int cnt = __be16_to_cpu(sb->phys->used_pdes);
1059                         for (j=0; j<cnt; j++)
1060                                 if (vc->phys_refnum[i] == sb->phys->entries[j].refnum)
1061                                         break;
1062                         if (i) printf(" ");
1063                         if (j < cnt)
1064                                 printf("%d", j);
1065                         else
1066                                 printf("--");
1067                 }
1068                 printf(")\n");
1069                 if (vc->chunk_shift != 255)
1070                 printf("   Chunk Size[%d] : %d sectors\n", n,
1071                        1 << vc->chunk_shift);
1072                 printf("   Raid Level[%d] : %s\n", n,
1073                        map_num(ddf_level, vc->prl)?:"-unknown-");
1074                 if (vc->sec_elmnt_count != 1) {
1075                         printf("  Secondary Position[%d] : %d of %d\n", n,
1076                                vc->sec_elmnt_seq, vc->sec_elmnt_count);
1077                         printf("  Secondary Level[%d] : %s\n", n,
1078                                map_num(ddf_sec_level, vc->srl) ?: "-unknown-");
1079                 }
1080                 printf("  Device Size[%d] : %llu\n", n,
1081                        (unsigned long long)__be64_to_cpu(vc->blocks)/2);
1082                 printf("   Array Size[%d] : %llu\n", n,
1083                        (unsigned long long)__be64_to_cpu(vc->array_blocks)/2);
1084         }
1085 }
1086
1087 static void examine_vds(struct ddf_super *sb)
1088 {
1089         int cnt = __be16_to_cpu(sb->virt->populated_vdes);
1090         int i;
1091         printf("  Virtual Disks : %d\n", cnt);
1092
1093         for (i=0; i<cnt; i++) {
1094                 struct virtual_entry *ve = &sb->virt->entries[i];
1095                 printf("\n");
1096                 printf("      VD GUID[%d] : ", i); print_guid(ve->guid, 1);
1097                 printf("\n");
1098                 printf("         unit[%d] : %d\n", i, __be16_to_cpu(ve->unit));
1099                 printf("        state[%d] : %s, %s%s\n", i,
1100                        map_num(ddf_state, ve->state & 7),
1101                        (ve->state & 8) ? "Morphing, ": "",
1102                        (ve->state & 16)? "Not Consistent" : "Consistent");
1103                 printf("   init state[%d] : %s\n", i,
1104                        map_num(ddf_init_state, ve->init_state&3));
1105                 printf("       access[%d] : %s\n", i,
1106                        map_num(ddf_access, (ve->init_state>>6) & 3));
1107                 printf("         Name[%d] : %.16s\n", i, ve->name);
1108                 examine_vd(i, sb, ve->guid);
1109         }
1110         if (cnt) printf("\n");
1111 }
1112
1113 static void examine_pds(struct ddf_super *sb)
1114 {
1115         int cnt = __be16_to_cpu(sb->phys->used_pdes);
1116         int i;
1117         struct dl *dl;
1118         printf(" Physical Disks : %d\n", cnt);
1119         printf("      Number    RefNo      Size       Device      Type/State\n");
1120
1121         for (i=0 ; i<cnt ; i++) {
1122                 struct phys_disk_entry *pd = &sb->phys->entries[i];
1123                 int type = __be16_to_cpu(pd->type);
1124                 int state = __be16_to_cpu(pd->state);
1125
1126                 //printf("      PD GUID[%d] : ", i); print_guid(pd->guid, 0);
1127                 //printf("\n");
1128                 printf("       %3d    %08x  ", i,
1129                        __be32_to_cpu(pd->refnum));
1130                 printf("%8lluK ", 
1131                        (unsigned long long)__be64_to_cpu(pd->config_size)>>1);
1132                 for (dl = sb->dlist; dl ; dl = dl->next) {
1133                         if (dl->disk.refnum == pd->refnum) {
1134                                 char *dv = map_dev(dl->major, dl->minor, 0);
1135                                 if (dv) {
1136                                         printf("%-15s", dv);
1137                                         break;
1138                                 }
1139                         }
1140                 }
1141                 if (!dl)
1142                         printf("%15s","");
1143                 printf(" %s%s%s%s%s",
1144                        (type&2) ? "active":"",
1145                        (type&4) ? "Global-Spare":"",
1146                        (type&8) ? "spare" : "",
1147                        (type&16)? ", foreign" : "",
1148                        (type&32)? "pass-through" : "");
1149                 printf("/%s%s%s%s%s%s%s",
1150                        (state&1)? "Online": "Offline",
1151                        (state&2)? ", Failed": "",
1152                        (state&4)? ", Rebuilding": "",
1153                        (state&8)? ", in-transition": "",
1154                        (state&16)? ", SMART-errors": "",
1155                        (state&32)? ", Unrecovered-Read-Errors": "",
1156                        (state&64)? ", Missing" : "");
1157                 printf("\n");
1158         }
1159 }
1160
1161 static void examine_super_ddf(struct supertype *st, char *homehost)
1162 {
1163         struct ddf_super *sb = st->sb;
1164
1165         printf("          Magic : %08x\n", __be32_to_cpu(sb->anchor.magic));
1166         printf("        Version : %.8s\n", sb->anchor.revision);
1167         printf("Controller GUID : "); print_guid(sb->controller.guid, 0);
1168         printf("\n");
1169         printf(" Container GUID : "); print_guid(sb->anchor.guid, 1);
1170         printf("\n");
1171         printf("            Seq : %08x\n", __be32_to_cpu(sb->active->seq));
1172         printf("  Redundant hdr : %s\n", sb->secondary.magic == DDF_HEADER_MAGIC
1173                ?"yes" : "no");
1174         examine_vds(sb);
1175         examine_pds(sb);
1176 }
1177
1178 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info);
1179
1180 static void uuid_from_super_ddf(struct supertype *st, int uuid[4]);
1181
1182 static void brief_examine_super_ddf(struct supertype *st, int verbose)
1183 {
1184         /* We just write a generic DDF ARRAY entry
1185          */
1186         struct ddf_super *ddf = st->sb;
1187         struct mdinfo info;
1188         int i;
1189         char nbuf[64];
1190         getinfo_super_ddf(st, &info);
1191         fname_from_uuid(st, &info, nbuf, ':');
1192
1193         for (i=0; i<__be16_to_cpu(ddf->virt->max_vdes); i++) {
1194                 struct virtual_entry *ve = &ddf->virt->entries[i];
1195                 struct vcl vcl;
1196                 char nbuf1[64];
1197                 if (all_ff(ve->guid))
1198                         continue;
1199                 memcpy(vcl.conf.guid, ve->guid, DDF_GUID_LEN);
1200                 ddf->currentconf =&vcl;
1201                 uuid_from_super_ddf(st, info.uuid);
1202                 fname_from_uuid(st, &info, nbuf1, ':');
1203                 printf("ARRAY container=%s member=%d UUID=%s\n",
1204                        nbuf+5, i, nbuf1+5);
1205         }
1206         printf("ARRAY metadata=ddf UUID=%s\n", nbuf + 5);
1207 }
1208
1209 static void export_examine_super_ddf(struct supertype *st)
1210 {
1211         struct mdinfo info;
1212         char nbuf[64];
1213         getinfo_super_ddf(st, &info);
1214         fname_from_uuid(st, &info, nbuf, ':');
1215         printf("MD_METADATA=ddf\n");
1216         printf("MD_LEVEL=container\n");
1217         printf("MD_UUID=%s\n", nbuf+5);
1218 }
1219         
1220
1221 static void detail_super_ddf(struct supertype *st, char *homehost)
1222 {
1223         /* FIXME later
1224          * Could print DDF GUID
1225          * Need to find which array
1226          *  If whole, briefly list all arrays
1227          *  If one, give name
1228          */
1229 }
1230
1231 static void brief_detail_super_ddf(struct supertype *st)
1232 {
1233         /* FIXME I really need to know which array we are detailing.
1234          * Can that be stored in ddf_super??
1235          */
1236 //      struct ddf_super *ddf = st->sb;
1237         struct mdinfo info;
1238         char nbuf[64];
1239         getinfo_super_ddf(st, &info);
1240         fname_from_uuid(st, &info, nbuf,':');
1241         printf(" UUID=%s", nbuf + 5);
1242 }
1243 #endif
1244
1245 static int match_home_ddf(struct supertype *st, char *homehost)
1246 {
1247         /* It matches 'this' host if the controller is a
1248          * Linux-MD controller with vendor_data matching
1249          * the hostname
1250          */
1251         struct ddf_super *ddf = st->sb;
1252         int len = strlen(homehost);
1253
1254         return (memcmp(ddf->controller.guid, T10, 8) == 0 &&
1255                 len < sizeof(ddf->controller.vendor_data) &&
1256                 memcmp(ddf->controller.vendor_data, homehost,len) == 0 &&
1257                 ddf->controller.vendor_data[len] == 0);
1258 }
1259
1260 #ifndef MDASSEMBLE
1261 static struct vd_config *find_vdcr(struct ddf_super *ddf, int inst)
1262 {
1263         struct vcl *v;
1264
1265         for (v = ddf->conflist; v; v = v->next)
1266                 if (inst == v->vcnum)
1267                         return &v->conf;
1268         return NULL;
1269 }
1270 #endif
1271
1272 static int find_phys(struct ddf_super *ddf, __u32 phys_refnum)
1273 {
1274         /* Find the entry in phys_disk which has the given refnum
1275          * and return it's index
1276          */
1277         int i;
1278         for (i=0; i < __be16_to_cpu(ddf->phys->max_pdes); i++)
1279                 if (ddf->phys->entries[i].refnum == phys_refnum)
1280                         return i;
1281         return -1;
1282 }
1283
1284 static void uuid_from_super_ddf(struct supertype *st, int uuid[4])
1285 {
1286         /* The uuid returned here is used for:
1287          *  uuid to put into bitmap file (Create, Grow)
1288          *  uuid for backup header when saving critical section (Grow)
1289          *  comparing uuids when re-adding a device into an array
1290          *    In these cases the uuid required is that of the data-array,
1291          *    not the device-set.
1292          *  uuid to recognise same set when adding a missing device back
1293          *    to an array.   This is a uuid for the device-set.
1294          *  
1295          * For each of these we can make do with a truncated
1296          * or hashed uuid rather than the original, as long as
1297          * everyone agrees.
1298          * In the case of SVD we assume the BVD is of interest,
1299          * though that might be the case if a bitmap were made for
1300          * a mirrored SVD - worry about that later.
1301          * So we need to find the VD configuration record for the
1302          * relevant BVD and extract the GUID and Secondary_Element_Seq.
1303          * The first 16 bytes of the sha1 of these is used.
1304          */
1305         struct ddf_super *ddf = st->sb;
1306         struct vcl *vcl = ddf->currentconf;
1307         char *guid;
1308         char buf[20];
1309         struct sha1_ctx ctx;
1310
1311         if (vcl)
1312                 guid = vcl->conf.guid;
1313         else
1314                 guid = ddf->anchor.guid;
1315
1316         sha1_init_ctx(&ctx);
1317         sha1_process_bytes(guid, DDF_GUID_LEN, &ctx);
1318         sha1_finish_ctx(&ctx, buf);
1319         memcpy(uuid, buf, 4*4);
1320 }
1321
1322 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info);
1323
1324 static void getinfo_super_ddf(struct supertype *st, struct mdinfo *info)
1325 {
1326         struct ddf_super *ddf = st->sb;
1327
1328         if (ddf->currentconf) {
1329                 getinfo_super_ddf_bvd(st, info);
1330                 return;
1331         }
1332
1333         info->array.raid_disks    = __be16_to_cpu(ddf->phys->used_pdes);
1334         info->array.level         = LEVEL_CONTAINER;
1335         info->array.layout        = 0;
1336         info->array.md_minor      = -1;
1337         info->array.ctime         = DECADE + __be32_to_cpu(*(__u32*)
1338                                                          (ddf->anchor.guid+16));
1339         info->array.utime         = 0;
1340         info->array.chunk_size    = 0;
1341
1342
1343         info->disk.major = 0;
1344         info->disk.minor = 0;
1345         if (ddf->dlist) {
1346                 info->disk.number = __be32_to_cpu(ddf->dlist->disk.refnum);
1347                 info->disk.raid_disk = find_phys(ddf, ddf->dlist->disk.refnum);
1348
1349                 info->data_offset = __be64_to_cpu(ddf->phys->
1350                                           entries[info->disk.raid_disk].
1351                                           config_size);
1352                 info->component_size = ddf->dlist->size - info->data_offset;
1353         } else {
1354                 info->disk.number = -1;
1355                 info->disk.raid_disk = -1;
1356 //              info->disk.raid_disk = find refnum in the table and use index;
1357         }
1358         info->disk.state = (1 << MD_DISK_SYNC);
1359
1360
1361         info->reshape_active = 0;
1362         info->name[0] = 0;
1363
1364         info->array.major_version = -1;
1365         info->array.minor_version = -2;
1366         strcpy(info->text_version, "ddf");
1367         info->safe_mode_delay = 0;
1368
1369         uuid_from_super_ddf(st, info->uuid);
1370
1371 }
1372
1373 static int rlq_to_layout(int rlq, int prl, int raiddisks);
1374
1375 static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
1376 {
1377         struct ddf_super *ddf = st->sb;
1378         struct vcl *vc = ddf->currentconf;
1379         int cd = ddf->currentdev;
1380         int j;
1381         struct dl *dl;
1382
1383         /* FIXME this returns BVD info - what if we want SVD ?? */
1384
1385         info->array.raid_disks    = __be16_to_cpu(vc->conf.prim_elmnt_count);
1386         info->array.level         = map_num1(ddf_level_num, vc->conf.prl);
1387         info->array.layout        = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
1388                                                   info->array.raid_disks);
1389         info->array.md_minor      = -1;
1390         info->array.ctime         = DECADE +
1391                 __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
1392         info->array.utime         = DECADE + __be32_to_cpu(vc->conf.timestamp);
1393         info->array.chunk_size    = 512 << vc->conf.chunk_shift;
1394         info->custom_array_size   = 0;
1395
1396         if (cd >= 0 && cd < ddf->mppe) {
1397                 info->data_offset         = __be64_to_cpu(vc->lba_offset[cd]);
1398                 if (vc->block_sizes)
1399                         info->component_size = vc->block_sizes[cd];
1400                 else
1401                         info->component_size = __be64_to_cpu(vc->conf.blocks);
1402         }
1403
1404         for (dl = ddf->dlist; dl ; dl = dl->next)
1405                 if (dl->raiddisk == info->disk.raid_disk)
1406                         break;
1407         info->disk.major = 0;
1408         info->disk.minor = 0;
1409         if (dl) {
1410                 info->disk.major = dl->major;
1411                 info->disk.minor = dl->minor;
1412         }
1413 //      info->disk.number = __be32_to_cpu(ddf->disk.refnum);
1414 //      info->disk.raid_disk = find refnum in the table and use index;
1415 //      info->disk.state = ???;
1416
1417         info->container_member = ddf->currentconf->vcnum;
1418
1419         info->resync_start = 0;
1420         if (!(ddf->virt->entries[info->container_member].state
1421               & DDF_state_inconsistent)  &&
1422             (ddf->virt->entries[info->container_member].init_state
1423              & DDF_initstate_mask)
1424             == DDF_init_full)
1425                 info->resync_start = ~0ULL;
1426
1427         uuid_from_super_ddf(st, info->uuid);
1428
1429         info->container_member = atoi(st->subarray);
1430         info->array.major_version = -1;
1431         info->array.minor_version = -2;
1432         sprintf(info->text_version, "/%s/%s",
1433                 devnum2devname(st->container_dev),
1434                 st->subarray);
1435         info->safe_mode_delay = 200;
1436
1437         memcpy(info->name, ddf->virt->entries[info->container_member].name, 16);
1438         info->name[16]=0;
1439         for(j=0; j<16; j++)
1440                 if (info->name[j] == ' ')
1441                         info->name[j] = 0;
1442 }
1443
1444
1445 static int update_super_ddf(struct supertype *st, struct mdinfo *info,
1446                             char *update,
1447                             char *devname, int verbose,
1448                             int uuid_set, char *homehost)
1449 {
1450         /* For 'assemble' and 'force' we need to return non-zero if any
1451          * change was made.  For others, the return value is ignored.
1452          * Update options are:
1453          *  force-one : This device looks a bit old but needs to be included,
1454          *        update age info appropriately.
1455          *  assemble: clear any 'faulty' flag to allow this device to
1456          *              be assembled.
1457          *  force-array: Array is degraded but being forced, mark it clean
1458          *         if that will be needed to assemble it.
1459          *
1460          *  newdev:  not used ????
1461          *  grow:  Array has gained a new device - this is currently for
1462          *              linear only
1463          *  resync: mark as dirty so a resync will happen.
1464          *  uuid:  Change the uuid of the array to match what is given
1465          *  homehost:  update the recorded homehost
1466          *  name:  update the name - preserving the homehost
1467          *  _reshape_progress: record new reshape_progress position.
1468          *
1469          * Following are not relevant for this version:
1470          *  sparc2.2 : update from old dodgey metadata
1471          *  super-minor: change the preferred_minor number
1472          *  summaries:  update redundant counters.
1473          */
1474         int rv = 0;
1475 //      struct ddf_super *ddf = st->sb;
1476 //      struct vd_config *vd = find_vdcr(ddf, info->container_member);
1477 //      struct virtual_entry *ve = find_ve(ddf);
1478
1479         /* we don't need to handle "force-*" or "assemble" as
1480          * there is no need to 'trick' the kernel.  We the metadata is
1481          * first updated to activate the array, all the implied modifications
1482          * will just happen.
1483          */
1484
1485         if (strcmp(update, "grow") == 0) {
1486                 /* FIXME */
1487         }
1488         if (strcmp(update, "resync") == 0) {
1489 //              info->resync_checkpoint = 0;
1490         }
1491         /* We ignore UUID updates as they make even less sense
1492          * with DDF
1493          */
1494         if (strcmp(update, "homehost") == 0) {
1495                 /* homehost is stored in controller->vendor_data,
1496                  * or it is when we are the vendor
1497                  */
1498 //              if (info->vendor_is_local)
1499 //                      strcpy(ddf->controller.vendor_data, homehost);
1500         }
1501         if (strcmp(update, "name") == 0) {
1502                 /* name is stored in virtual_entry->name */
1503 //              memset(ve->name, ' ', 16);
1504 //              strncpy(ve->name, info->name, 16);
1505         }
1506         if (strcmp(update, "_reshape_progress") == 0) {
1507                 /* We don't support reshape yet */
1508         }
1509
1510 //      update_all_csum(ddf);
1511
1512         return rv;
1513 }
1514
1515 static void make_header_guid(char *guid)
1516 {
1517         __u32 stamp;
1518         /* Create a DDF Header of Virtual Disk GUID */
1519
1520         /* 24 bytes of fiction required.
1521          * first 8 are a 'vendor-id'  - "Linux-MD"
1522          * next 8 are controller type.. how about 0X DEAD BEEF 0000 0000
1523          * Remaining 8 random number plus timestamp
1524          */
1525         memcpy(guid, T10, sizeof(T10));
1526         stamp = __cpu_to_be32(0xdeadbeef);
1527         memcpy(guid+8, &stamp, 4);
1528         stamp = __cpu_to_be32(0);
1529         memcpy(guid+12, &stamp, 4);
1530         stamp = __cpu_to_be32(time(0) - DECADE);
1531         memcpy(guid+16, &stamp, 4);
1532         stamp = random32();
1533         memcpy(guid+20, &stamp, 4);
1534 }
1535
1536 static int init_super_ddf_bvd(struct supertype *st,
1537                               mdu_array_info_t *info,
1538                               unsigned long long size,
1539                               char *name, char *homehost,
1540                               int *uuid);
1541
1542 static int init_super_ddf(struct supertype *st,
1543                           mdu_array_info_t *info,
1544                           unsigned long long size, char *name, char *homehost,
1545                           int *uuid)
1546 {
1547         /* This is primarily called by Create when creating a new array.
1548          * We will then get add_to_super called for each component, and then
1549          * write_init_super called to write it out to each device.
1550          * For DDF, Create can create on fresh devices or on a pre-existing
1551          * array.
1552          * To create on a pre-existing array a different method will be called.
1553          * This one is just for fresh drives.
1554          *
1555          * We need to create the entire 'ddf' structure which includes:
1556          *  DDF headers - these are easy.
1557          *  Controller data - a Sector describing this controller .. not that
1558          *                  this is a controller exactly.
1559          *  Physical Disk Record - one entry per device, so
1560          *                      leave plenty of space.
1561          *  Virtual Disk Records - again, just leave plenty of space.
1562          *                   This just lists VDs, doesn't give details
1563          *  Config records - describes the VDs that use this disk
1564          *  DiskData  - describes 'this' device.
1565          *  BadBlockManagement - empty
1566          *  Diag Space - empty
1567          *  Vendor Logs - Could we put bitmaps here?
1568          *
1569          */
1570         struct ddf_super *ddf;
1571         char hostname[17];
1572         int hostlen;
1573         int max_phys_disks, max_virt_disks;
1574         unsigned long long sector;
1575         int clen;
1576         int i;
1577         int pdsize, vdsize;
1578         struct phys_disk *pd;
1579         struct virtual_disk *vd;
1580
1581         if (!info) {
1582                 st->sb = NULL;
1583                 return 0;
1584         }
1585         if (st->sb)
1586                 return init_super_ddf_bvd(st, info, size, name, homehost,
1587                                           uuid);
1588
1589         if (posix_memalign((void**)&ddf, 512, sizeof(*ddf)) != 0) {
1590                 fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
1591                 return 0;
1592         }
1593         memset(ddf, 0, sizeof(*ddf));
1594         ddf->dlist = NULL; /* no physical disks yet */
1595         ddf->conflist = NULL; /* No virtual disks yet */
1596
1597         /* At least 32MB *must* be reserved for the ddf.  So let's just
1598          * start 32MB from the end, and put the primary header there.
1599          * Don't do secondary for now.
1600          * We don't know exactly where that will be yet as it could be
1601          * different on each device.  To just set up the lengths.
1602          *
1603          */
1604
1605         ddf->anchor.magic = DDF_HEADER_MAGIC;
1606         make_header_guid(ddf->anchor.guid);
1607
1608         memcpy(ddf->anchor.revision, DDF_REVISION_2, 8);
1609         ddf->anchor.seq = __cpu_to_be32(1);
1610         ddf->anchor.timestamp = __cpu_to_be32(time(0) - DECADE);
1611         ddf->anchor.openflag = 0xFF;
1612         ddf->anchor.foreignflag = 0;
1613         ddf->anchor.enforcegroups = 0; /* Is this best?? */
1614         ddf->anchor.pad0 = 0xff;
1615         memset(ddf->anchor.pad1, 0xff, 12);
1616         memset(ddf->anchor.header_ext, 0xff, 32);
1617         ddf->anchor.primary_lba = ~(__u64)0;
1618         ddf->anchor.secondary_lba = ~(__u64)0;
1619         ddf->anchor.type = DDF_HEADER_ANCHOR;
1620         memset(ddf->anchor.pad2, 0xff, 3);
1621         ddf->anchor.workspace_len = __cpu_to_be32(32768); /* Must be reserved */
1622         ddf->anchor.workspace_lba = ~(__u64)0; /* Put this at bottom
1623                                                   of 32M reserved.. */
1624         max_phys_disks = 1023;   /* Should be enough */
1625         ddf->anchor.max_pd_entries = __cpu_to_be16(max_phys_disks);
1626         max_virt_disks = 255;
1627         ddf->anchor.max_vd_entries = __cpu_to_be16(max_virt_disks); /* ?? */
1628         ddf->anchor.max_partitions = __cpu_to_be16(64); /* ?? */
1629         ddf->max_part = 64;
1630         ddf->mppe = 256;
1631         ddf->conf_rec_len = 1 + ROUND_UP(ddf->mppe * (4+8), 512)/512;
1632         ddf->anchor.config_record_len = __cpu_to_be16(ddf->conf_rec_len);
1633         ddf->anchor.max_primary_element_entries = __cpu_to_be16(ddf->mppe);
1634         memset(ddf->anchor.pad3, 0xff, 54);
1635         /* controller sections is one sector long immediately
1636          * after the ddf header */
1637         sector = 1;
1638         ddf->anchor.controller_section_offset = __cpu_to_be32(sector);
1639         ddf->anchor.controller_section_length = __cpu_to_be32(1);
1640         sector += 1;
1641
1642         /* phys is 8 sectors after that */
1643         pdsize = ROUND_UP(sizeof(struct phys_disk) +
1644                           sizeof(struct phys_disk_entry)*max_phys_disks,
1645                           512);
1646         switch(pdsize/512) {
1647         case 2: case 8: case 32: case 128: case 512: break;
1648         default: abort();
1649         }
1650         ddf->anchor.phys_section_offset = __cpu_to_be32(sector);
1651         ddf->anchor.phys_section_length =
1652                 __cpu_to_be32(pdsize/512); /* max_primary_element_entries/8 */
1653         sector += pdsize/512;
1654
1655         /* virt is another 32 sectors */
1656         vdsize = ROUND_UP(sizeof(struct virtual_disk) +
1657                           sizeof(struct virtual_entry) * max_virt_disks,
1658                           512);
1659         switch(vdsize/512) {
1660         case 2: case 8: case 32: case 128: case 512: break;
1661         default: abort();
1662         }
1663         ddf->anchor.virt_section_offset = __cpu_to_be32(sector);
1664         ddf->anchor.virt_section_length =
1665                 __cpu_to_be32(vdsize/512); /* max_vd_entries/8 */
1666         sector += vdsize/512;
1667
1668         clen = ddf->conf_rec_len * (ddf->max_part+1);
1669         ddf->anchor.config_section_offset = __cpu_to_be32(sector);
1670         ddf->anchor.config_section_length = __cpu_to_be32(clen);
1671         sector += clen;
1672
1673         ddf->anchor.data_section_offset = __cpu_to_be32(sector);
1674         ddf->anchor.data_section_length = __cpu_to_be32(1);
1675         sector += 1;
1676
1677         ddf->anchor.bbm_section_length = __cpu_to_be32(0);
1678         ddf->anchor.bbm_section_offset = __cpu_to_be32(0xFFFFFFFF);
1679         ddf->anchor.diag_space_length = __cpu_to_be32(0);
1680         ddf->anchor.diag_space_offset = __cpu_to_be32(0xFFFFFFFF);
1681         ddf->anchor.vendor_length = __cpu_to_be32(0);
1682         ddf->anchor.vendor_offset = __cpu_to_be32(0xFFFFFFFF);
1683
1684         memset(ddf->anchor.pad4, 0xff, 256);
1685
1686         memcpy(&ddf->primary, &ddf->anchor, 512);
1687         memcpy(&ddf->secondary, &ddf->anchor, 512);
1688
1689         ddf->primary.openflag = 1; /* I guess.. */
1690         ddf->primary.type = DDF_HEADER_PRIMARY;
1691
1692         ddf->secondary.openflag = 1; /* I guess.. */
1693         ddf->secondary.type = DDF_HEADER_SECONDARY;
1694
1695         ddf->active = &ddf->primary;
1696
1697         ddf->controller.magic = DDF_CONTROLLER_MAGIC;
1698
1699         /* 24 more bytes of fiction required.
1700          * first 8 are a 'vendor-id'  - "Linux-MD"
1701          * Remaining 16 are serial number.... maybe a hostname would do?
1702          */
1703         memcpy(ddf->controller.guid, T10, sizeof(T10));
1704         gethostname(hostname, sizeof(hostname));
1705         hostname[sizeof(hostname) - 1] = 0;
1706         hostlen = strlen(hostname);
1707         memcpy(ddf->controller.guid + 24 - hostlen, hostname, hostlen);
1708         for (i = strlen(T10) ; i+hostlen < 24; i++)
1709                 ddf->controller.guid[i] = ' ';
1710
1711         ddf->controller.type.vendor_id = __cpu_to_be16(0xDEAD);
1712         ddf->controller.type.device_id = __cpu_to_be16(0xBEEF);
1713         ddf->controller.type.sub_vendor_id = 0;
1714         ddf->controller.type.sub_device_id = 0;
1715         memcpy(ddf->controller.product_id, "What Is My PID??", 16);
1716         memset(ddf->controller.pad, 0xff, 8);
1717         memset(ddf->controller.vendor_data, 0xff, 448);
1718         if (homehost && strlen(homehost) < 440)
1719                 strcpy((char*)ddf->controller.vendor_data, homehost);
1720
1721         if (posix_memalign((void**)&pd, 512, pdsize) != 0) {
1722                 fprintf(stderr, Name ": %s could not allocate pd\n", __func__);
1723                 return 0;
1724         }
1725         ddf->phys = pd;
1726         ddf->pdsize = pdsize;
1727
1728         memset(pd, 0xff, pdsize);
1729         memset(pd, 0, sizeof(*pd));
1730         pd->magic = DDF_PHYS_DATA_MAGIC;
1731         pd->used_pdes = __cpu_to_be16(0);
1732         pd->max_pdes = __cpu_to_be16(max_phys_disks);
1733         memset(pd->pad, 0xff, 52);
1734
1735         if (posix_memalign((void**)&vd, 512, vdsize) != 0) {
1736                 fprintf(stderr, Name ": %s could not allocate vd\n", __func__);
1737                 return 0;
1738         }
1739         ddf->virt = vd;
1740         ddf->vdsize = vdsize;
1741         memset(vd, 0, vdsize);
1742         vd->magic = DDF_VIRT_RECORDS_MAGIC;
1743         vd->populated_vdes = __cpu_to_be16(0);
1744         vd->max_vdes = __cpu_to_be16(max_virt_disks);
1745         memset(vd->pad, 0xff, 52);
1746
1747         for (i=0; i<max_virt_disks; i++)
1748                 memset(&vd->entries[i], 0xff, sizeof(struct virtual_entry));
1749
1750         st->sb = ddf;
1751         ddf->updates_pending = 1;
1752         return 1;
1753 }
1754
1755 static int chunk_to_shift(int chunksize)
1756 {
1757         return ffs(chunksize/512)-1;
1758 }
1759
1760 static int level_to_prl(int level)
1761 {
1762         switch (level) {
1763         case LEVEL_LINEAR: return DDF_CONCAT;
1764         case 0: return DDF_RAID0;
1765         case 1: return DDF_RAID1;
1766         case 4: return DDF_RAID4;
1767         case 5: return DDF_RAID5;
1768         case 6: return DDF_RAID6;
1769         default: return -1;
1770         }
1771 }
1772 static int layout_to_rlq(int level, int layout, int raiddisks)
1773 {
1774         switch(level) {
1775         case 0:
1776                 return DDF_RAID0_SIMPLE;
1777         case 1:
1778                 switch(raiddisks) {
1779                 case 2: return DDF_RAID1_SIMPLE;
1780                 case 3: return DDF_RAID1_MULTI;
1781                 default: return -1;
1782                 }
1783         case 4:
1784                 switch(layout) {
1785                 case 0: return DDF_RAID4_N;
1786                 }
1787                 break;
1788         case 5:
1789                 switch(layout) {
1790                 case ALGORITHM_LEFT_ASYMMETRIC:
1791                         return DDF_RAID5_N_RESTART;
1792                 case ALGORITHM_RIGHT_ASYMMETRIC:
1793                         return DDF_RAID5_0_RESTART;
1794                 case ALGORITHM_LEFT_SYMMETRIC:
1795                         return DDF_RAID5_N_CONTINUE;
1796                 case ALGORITHM_RIGHT_SYMMETRIC:
1797                         return -1; /* not mentioned in standard */
1798                 }
1799         case 6:
1800                 switch(layout) {
1801                 case ALGORITHM_ROTATING_N_RESTART:
1802                         return DDF_RAID5_N_RESTART;
1803                 case ALGORITHM_ROTATING_ZERO_RESTART:
1804                         return DDF_RAID6_0_RESTART;
1805                 case ALGORITHM_ROTATING_N_CONTINUE:
1806                         return DDF_RAID5_N_CONTINUE;
1807                 }
1808         }
1809         return -1;
1810 }
1811
1812 static int rlq_to_layout(int rlq, int prl, int raiddisks)
1813 {
1814         switch(prl) {
1815         case DDF_RAID0:
1816                 return 0; /* hopefully rlq == DDF_RAID0_SIMPLE */
1817         case DDF_RAID1:
1818                 return 0; /* hopefully rlq == SIMPLE or MULTI depending
1819                              on raiddisks*/
1820         case DDF_RAID4:
1821                 switch(rlq) {
1822                 case DDF_RAID4_N:
1823                         return 0;
1824                 default:
1825                         /* not supported */
1826                         return -1; /* FIXME this isn't checked */
1827                 }
1828         case DDF_RAID5:
1829                 switch(rlq) {
1830                 case DDF_RAID5_N_RESTART:
1831                         return ALGORITHM_LEFT_ASYMMETRIC;
1832                 case DDF_RAID5_0_RESTART:
1833                         return ALGORITHM_RIGHT_ASYMMETRIC;
1834                 case DDF_RAID5_N_CONTINUE:
1835                         return ALGORITHM_LEFT_SYMMETRIC;
1836                 default:
1837                         return -1;
1838                 }
1839         case DDF_RAID6:
1840                 switch(rlq) {
1841                 case DDF_RAID5_N_RESTART:
1842                         return ALGORITHM_ROTATING_N_RESTART;
1843                 case DDF_RAID6_0_RESTART:
1844                         return ALGORITHM_ROTATING_ZERO_RESTART;
1845                 case DDF_RAID5_N_CONTINUE:
1846                         return ALGORITHM_ROTATING_N_CONTINUE;
1847                 default:
1848                         return -1;
1849                 }
1850         }
1851         return -1;
1852 }
1853
1854 #ifndef MDASSEMBLE
1855 struct extent {
1856         unsigned long long start, size;
1857 };
1858 static int cmp_extent(const void *av, const void *bv)
1859 {
1860         const struct extent *a = av;
1861         const struct extent *b = bv;
1862         if (a->start < b->start)
1863                 return -1;
1864         if (a->start > b->start)
1865                 return 1;
1866         return 0;
1867 }
1868
1869 static struct extent *get_extents(struct ddf_super *ddf, struct dl *dl)
1870 {
1871         /* find a list of used extents on the give physical device
1872          * (dnum) of the given ddf.
1873          * Return a malloced array of 'struct extent'
1874
1875 FIXME ignore DDF_Legacy devices?
1876
1877          */
1878         struct extent *rv;
1879         int n = 0;
1880         int i, j;
1881
1882         rv = malloc(sizeof(struct extent) * (ddf->max_part + 2));
1883         if (!rv)
1884                 return NULL;
1885
1886         for (i = 0; i < ddf->max_part; i++) {
1887                 struct vcl *v = dl->vlist[i];
1888                 if (v == NULL)
1889                         continue;
1890                 for (j=0; j < v->conf.prim_elmnt_count; j++)
1891                         if (v->conf.phys_refnum[j] == dl->disk.refnum) {
1892                                 /* This device plays role 'j' in  'v'. */
1893                                 rv[n].start = __be64_to_cpu(v->lba_offset[j]);
1894                                 rv[n].size = __be64_to_cpu(v->conf.blocks);
1895                                 n++;
1896                                 break;
1897                         }
1898         }
1899         qsort(rv, n, sizeof(*rv), cmp_extent);
1900
1901         rv[n].start = __be64_to_cpu(ddf->phys->entries[dl->pdnum].config_size);
1902         rv[n].size = 0;
1903         return rv;
1904 }
1905 #endif
1906
1907 static int init_super_ddf_bvd(struct supertype *st,
1908                               mdu_array_info_t *info,
1909                               unsigned long long size,
1910                               char *name, char *homehost,
1911                               int *uuid)
1912 {
1913         /* We are creating a BVD inside a pre-existing container.
1914          * so st->sb is already set.
1915          * We need to create a new vd_config and a new virtual_entry
1916          */
1917         struct ddf_super *ddf = st->sb;
1918         int venum;
1919         struct virtual_entry *ve;
1920         struct vcl *vcl;
1921         struct vd_config *vc;
1922
1923         if (__be16_to_cpu(ddf->virt->populated_vdes)
1924             >= __be16_to_cpu(ddf->virt->max_vdes)) {
1925                 fprintf(stderr, Name": This ddf already has the "
1926                         "maximum of %d virtual devices\n",
1927                         __be16_to_cpu(ddf->virt->max_vdes));
1928                 return 0;
1929         }
1930
1931         for (venum = 0; venum < __be16_to_cpu(ddf->virt->max_vdes); venum++)
1932                 if (all_ff(ddf->virt->entries[venum].guid))
1933                         break;
1934         if (venum == __be16_to_cpu(ddf->virt->max_vdes)) {
1935                 fprintf(stderr, Name ": Cannot find spare slot for "
1936                         "virtual disk - DDF is corrupt\n");
1937                 return 0;
1938         }
1939         ve = &ddf->virt->entries[venum];
1940
1941         /* A Virtual Disk GUID contains the T10 Vendor ID, controller type,
1942          * timestamp, random number
1943          */
1944         make_header_guid(ve->guid);
1945         ve->unit = __cpu_to_be16(info->md_minor);
1946         ve->pad0 = 0xFFFF;
1947         ve->guid_crc = crc32(0, (unsigned char*)ddf->anchor.guid, DDF_GUID_LEN);
1948         ve->type = 0;
1949         ve->state = DDF_state_degraded; /* Will be modified as devices are added */
1950         if (info->state & 1) /* clean */
1951                 ve->init_state = DDF_init_full;
1952         else
1953                 ve->init_state = DDF_init_not;
1954
1955         memset(ve->pad1, 0xff, 14);
1956         memset(ve->name, ' ', 16);
1957         if (name)
1958                 strncpy(ve->name, name, 16);
1959         ddf->virt->populated_vdes =
1960                 __cpu_to_be16(__be16_to_cpu(ddf->virt->populated_vdes)+1);
1961
1962         /* Now create a new vd_config */
1963         if (posix_memalign((void**)&vcl, 512,
1964                            (offsetof(struct vcl, conf) + ddf->conf_rec_len * 512)) != 0) {
1965                 fprintf(stderr, Name ": %s could not allocate vd_config\n", __func__);
1966                 return 0;
1967         }
1968         vcl->lba_offset = (__u64*) &vcl->conf.phys_refnum[ddf->mppe];
1969         vcl->vcnum = venum;
1970         sprintf(st->subarray, "%d", venum);
1971         vcl->block_sizes = NULL; /* FIXME not for CONCAT */
1972
1973         vc = &vcl->conf;
1974
1975         vc->magic = DDF_VD_CONF_MAGIC;
1976         memcpy(vc->guid, ve->guid, DDF_GUID_LEN);
1977         vc->timestamp = __cpu_to_be32(time(0)-DECADE);
1978         vc->seqnum = __cpu_to_be32(1);
1979         memset(vc->pad0, 0xff, 24);
1980         vc->prim_elmnt_count = __cpu_to_be16(info->raid_disks);
1981         vc->chunk_shift = chunk_to_shift(info->chunk_size);
1982         vc->prl = level_to_prl(info->level);
1983         vc->rlq = layout_to_rlq(info->level, info->layout, info->raid_disks);
1984         vc->sec_elmnt_count = 1;
1985         vc->sec_elmnt_seq = 0;
1986         vc->srl = 0;
1987         vc->blocks = __cpu_to_be64(info->size * 2);
1988         vc->array_blocks = __cpu_to_be64(
1989                 calc_array_size(info->level, info->raid_disks, info->layout,
1990                                 info->chunk_size, info->size*2));
1991         memset(vc->pad1, 0xff, 8);
1992         vc->spare_refs[0] = 0xffffffff;
1993         vc->spare_refs[1] = 0xffffffff;
1994         vc->spare_refs[2] = 0xffffffff;
1995         vc->spare_refs[3] = 0xffffffff;
1996         vc->spare_refs[4] = 0xffffffff;
1997         vc->spare_refs[5] = 0xffffffff;
1998         vc->spare_refs[6] = 0xffffffff;
1999         vc->spare_refs[7] = 0xffffffff;
2000         memset(vc->cache_pol, 0, 8);
2001         vc->bg_rate = 0x80;
2002         memset(vc->pad2, 0xff, 3);
2003         memset(vc->pad3, 0xff, 52);
2004         memset(vc->pad4, 0xff, 192);
2005         memset(vc->v0, 0xff, 32);
2006         memset(vc->v1, 0xff, 32);
2007         memset(vc->v2, 0xff, 16);
2008         memset(vc->v3, 0xff, 16);
2009         memset(vc->vendor, 0xff, 32);
2010
2011         memset(vc->phys_refnum, 0xff, 4*ddf->mppe);
2012         memset(vc->phys_refnum+ddf->mppe, 0x00, 8*ddf->mppe);
2013
2014         vcl->next = ddf->conflist;
2015         ddf->conflist = vcl;
2016         ddf->currentconf = vcl;
2017         ddf->updates_pending = 1;
2018         return 1;
2019 }
2020
2021 #ifndef MDASSEMBLE
2022 static void add_to_super_ddf_bvd(struct supertype *st,
2023                                  mdu_disk_info_t *dk, int fd, char *devname)
2024 {
2025         /* fd and devname identify a device with-in the ddf container (st).
2026          * dk identifies a location in the new BVD.
2027          * We need to find suitable free space in that device and update
2028          * the phys_refnum and lba_offset for the newly created vd_config.
2029          * We might also want to update the type in the phys_disk
2030          * section.
2031          *
2032          * Alternately: fd == -1 and we have already chosen which device to
2033          * use and recorded in dlist->raid_disk;
2034          */
2035         struct dl *dl;
2036         struct ddf_super *ddf = st->sb;
2037         struct vd_config *vc;
2038         __u64 *lba_offset;
2039         int working;
2040         int i;
2041         unsigned long long blocks, pos, esize;
2042         struct extent *ex;
2043
2044         if (fd == -1) {
2045                 for (dl = ddf->dlist; dl ; dl = dl->next)
2046                         if (dl->raiddisk == dk->raid_disk)
2047                                 break;
2048         } else {
2049                 for (dl = ddf->dlist; dl ; dl = dl->next)
2050                         if (dl->major == dk->major &&
2051                             dl->minor == dk->minor)
2052                                 break;
2053         }
2054         if (!dl || ! (dk->state & (1<<MD_DISK_SYNC)))
2055                 return;
2056
2057         vc = &ddf->currentconf->conf;
2058         lba_offset = ddf->currentconf->lba_offset;
2059
2060         ex = get_extents(ddf, dl);
2061         if (!ex)
2062                 return;
2063
2064         i = 0; pos = 0;
2065         blocks = __be64_to_cpu(vc->blocks);
2066         if (ddf->currentconf->block_sizes)
2067                 blocks = ddf->currentconf->block_sizes[dk->raid_disk];
2068
2069         do {
2070                 esize = ex[i].start - pos;
2071                 if (esize >= blocks)
2072                         break;
2073                 pos = ex[i].start + ex[i].size;
2074                 i++;
2075         } while (ex[i-1].size);
2076
2077         free(ex);
2078         if (esize < blocks)
2079                 return;
2080
2081         ddf->currentdev = dk->raid_disk;
2082         vc->phys_refnum[dk->raid_disk] = dl->disk.refnum;
2083         lba_offset[dk->raid_disk] = __cpu_to_be64(pos);
2084
2085         for (i=0; i < ddf->max_part ; i++)
2086                 if (dl->vlist[i] == NULL)
2087                         break;
2088         if (i == ddf->max_part)
2089                 return;
2090         dl->vlist[i] = ddf->currentconf;
2091
2092         if (fd >= 0)
2093                 dl->fd = fd;
2094         if (devname)
2095                 dl->devname = devname;
2096
2097         /* Check how many working raid_disks, and if we can mark
2098          * array as optimal yet
2099          */
2100         working = 0;
2101
2102         for (i=0; i < __be16_to_cpu(vc->prim_elmnt_count); i++)
2103                 if (vc->phys_refnum[i] != 0xffffffff)
2104                         working++;
2105
2106         /* Find which virtual_entry */
2107         i = ddf->currentconf->vcnum;
2108         if (working == __be16_to_cpu(vc->prim_elmnt_count))
2109                 ddf->virt->entries[i].state =
2110                         (ddf->virt->entries[i].state & ~DDF_state_mask)
2111                         | DDF_state_optimal;
2112
2113         if (vc->prl == DDF_RAID6 &&
2114             working+1 == __be16_to_cpu(vc->prim_elmnt_count))
2115                 ddf->virt->entries[i].state =
2116                         (ddf->virt->entries[i].state & ~DDF_state_mask)
2117                         | DDF_state_part_optimal;
2118
2119         ddf->phys->entries[dl->pdnum].type &= ~__cpu_to_be16(DDF_Global_Spare);
2120         ddf->phys->entries[dl->pdnum].type |= __cpu_to_be16(DDF_Active_in_VD);
2121         ddf->updates_pending = 1;
2122 }
2123
2124 /* add a device to a container, either while creating it or while
2125  * expanding a pre-existing container
2126  */
2127 static int add_to_super_ddf(struct supertype *st,
2128                              mdu_disk_info_t *dk, int fd, char *devname)
2129 {
2130         struct ddf_super *ddf = st->sb;
2131         struct dl *dd;
2132         time_t now;
2133         struct tm *tm;
2134         unsigned long long size;
2135         struct phys_disk_entry *pde;
2136         int n, i;
2137         struct stat stb;
2138
2139         if (ddf->currentconf) {
2140                 add_to_super_ddf_bvd(st, dk, fd, devname);
2141                 return 0;
2142         }
2143
2144         /* This is device numbered dk->number.  We need to create
2145          * a phys_disk entry and a more detailed disk_data entry.
2146          */
2147         fstat(fd, &stb);
2148         if (posix_memalign((void**)&dd, 512,
2149                            sizeof(*dd) + sizeof(dd->vlist[0]) * ddf->max_part) != 0) {
2150                 fprintf(stderr, Name
2151                         ": %s could allocate buffer for new disk, aborting\n",
2152                         __func__);
2153                 return 1;
2154         }
2155         dd->major = major(stb.st_rdev);
2156         dd->minor = minor(stb.st_rdev);
2157         dd->devname = devname;
2158         dd->fd = fd;
2159         dd->spare = NULL;
2160
2161         dd->disk.magic = DDF_PHYS_DATA_MAGIC;
2162         now = time(0);
2163         tm = localtime(&now);
2164         sprintf(dd->disk.guid, "%8s%04d%02d%02d",
2165                 T10, tm->tm_year+1900, tm->tm_mon+1, tm->tm_mday);
2166         *(__u32*)(dd->disk.guid + 16) = random32();
2167         *(__u32*)(dd->disk.guid + 20) = random32();
2168
2169         do {
2170                 /* Cannot be bothered finding a CRC of some irrelevant details*/
2171                 dd->disk.refnum = random32();
2172                 for (i = __be16_to_cpu(ddf->active->max_pd_entries) - 1;
2173                      i >= 0; i--)
2174                         if (ddf->phys->entries[i].refnum == dd->disk.refnum)
2175                                 break;
2176         } while (i >= 0);
2177
2178         dd->disk.forced_ref = 1;
2179         dd->disk.forced_guid = 1;
2180         memset(dd->disk.vendor, ' ', 32);
2181         memcpy(dd->disk.vendor, "Linux", 5);
2182         memset(dd->disk.pad, 0xff, 442);
2183         for (i = 0; i < ddf->max_part ; i++)
2184                 dd->vlist[i] = NULL;
2185
2186         n = __be16_to_cpu(ddf->phys->used_pdes);
2187         pde = &ddf->phys->entries[n];
2188         dd->pdnum = n;
2189
2190         if (st->update_tail) {
2191                 int len = (sizeof(struct phys_disk) +
2192                            sizeof(struct phys_disk_entry));
2193                 struct phys_disk *pd;
2194
2195                 pd = malloc(len);
2196                 pd->magic = DDF_PHYS_RECORDS_MAGIC;
2197                 pd->used_pdes = __cpu_to_be16(n);
2198                 pde = &pd->entries[0];
2199                 dd->mdupdate = pd;
2200         } else {
2201                 n++;
2202                 ddf->phys->used_pdes = __cpu_to_be16(n);
2203         }
2204
2205         memcpy(pde->guid, dd->disk.guid, DDF_GUID_LEN);
2206         pde->refnum = dd->disk.refnum;
2207         pde->type = __cpu_to_be16(DDF_Forced_PD_GUID | DDF_Global_Spare);
2208         pde->state = __cpu_to_be16(DDF_Online);
2209         get_dev_size(fd, NULL, &size);
2210         /* We are required to reserve 32Meg, and record the size in sectors */
2211         pde->config_size = __cpu_to_be64( (size - 32*1024*1024) / 512);
2212         sprintf(pde->path, "%17.17s","Information: nil") ;
2213         memset(pde->pad, 0xff, 6);
2214
2215         dd->size = size >> 9;
2216         if (st->update_tail) {
2217                 dd->next = ddf->add_list;
2218                 ddf->add_list = dd;
2219         } else {
2220                 dd->next = ddf->dlist;
2221                 ddf->dlist = dd;
2222                 ddf->updates_pending = 1;
2223         }
2224
2225         return 0;
2226 }
2227
2228 /*
2229  * This is the write_init_super method for a ddf container.  It is
2230  * called when creating a container or adding another device to a
2231  * container.
2232  */
2233
2234 static unsigned char null_conf[4096+512];
2235
2236 static int __write_init_super_ddf(struct supertype *st, int do_close)
2237 {
2238
2239         struct ddf_super *ddf = st->sb;
2240         int i;
2241         struct dl *d;
2242         int n_config;
2243         int conf_size;
2244         int attempts = 0;
2245         int successes = 0;
2246         unsigned long long size, sector;
2247
2248         /* try to write updated metadata,
2249          * if we catch a failure move on to the next disk
2250          */
2251         for (d = ddf->dlist; d; d=d->next) {
2252                 int fd = d->fd;
2253
2254                 if (fd < 0)
2255                         continue;
2256
2257                 attempts++;
2258                 /* We need to fill in the primary, (secondary) and workspace
2259                  * lba's in the headers, set their checksums,
2260                  * Also checksum phys, virt....
2261                  *
2262                  * Then write everything out, finally the anchor is written.
2263                  */
2264                 get_dev_size(fd, NULL, &size);
2265                 size /= 512;
2266                 ddf->anchor.workspace_lba = __cpu_to_be64(size - 32*1024*2);
2267                 ddf->anchor.primary_lba = __cpu_to_be64(size - 16*1024*2);
2268                 ddf->anchor.seq = __cpu_to_be32(1);
2269                 memcpy(&ddf->primary, &ddf->anchor, 512);
2270                 memcpy(&ddf->secondary, &ddf->anchor, 512);
2271
2272                 ddf->anchor.openflag = 0xFF; /* 'open' means nothing */
2273                 ddf->anchor.seq = 0xFFFFFFFF; /* no sequencing in anchor */
2274                 ddf->anchor.crc = calc_crc(&ddf->anchor, 512);
2275
2276                 ddf->primary.openflag = 0;
2277                 ddf->primary.type = DDF_HEADER_PRIMARY;
2278
2279                 ddf->secondary.openflag = 0;
2280                 ddf->secondary.type = DDF_HEADER_SECONDARY;
2281
2282                 ddf->primary.crc = calc_crc(&ddf->primary, 512);
2283                 ddf->secondary.crc = calc_crc(&ddf->secondary, 512);
2284
2285                 sector = size - 16*1024*2;
2286                 lseek64(fd, sector<<9, 0);
2287                 if (write(fd, &ddf->primary, 512) < 0)
2288                         continue;
2289
2290                 ddf->controller.crc = calc_crc(&ddf->controller, 512);
2291                 if (write(fd, &ddf->controller, 512) < 0)
2292                         continue;
2293
2294                 ddf->phys->crc = calc_crc(ddf->phys, ddf->pdsize);
2295
2296                 if (write(fd, ddf->phys, ddf->pdsize) < 0)
2297                         continue;
2298
2299                 ddf->virt->crc = calc_crc(ddf->virt, ddf->vdsize);
2300                 if (write(fd, ddf->virt, ddf->vdsize) < 0)
2301                         continue;
2302
2303                 /* Now write lots of config records. */
2304                 n_config = ddf->max_part;
2305                 conf_size = ddf->conf_rec_len * 512;
2306                 for (i = 0 ; i <= n_config ; i++) {
2307                         struct vcl *c = d->vlist[i];
2308                         if (i == n_config)
2309                                 c = (struct vcl*)d->spare;
2310
2311                         if (c) {
2312                                 c->conf.crc = calc_crc(&c->conf, conf_size);
2313                                 if (write(fd, &c->conf, conf_size) < 0)
2314                                         break;
2315                         } else {
2316                                 char *null_aligned = (char*)((((unsigned long)null_conf)+511)&~511UL);
2317                                 if (null_conf[0] != 0xff)
2318                                         memset(null_conf, 0xff, sizeof(null_conf));
2319                                 int togo = conf_size;
2320                                 while (togo > sizeof(null_conf)-512) {
2321                                         if (write(fd, null_aligned, sizeof(null_conf)-512) < 0)
2322                                                 break;
2323                                         togo -= sizeof(null_conf)-512;
2324                                 }
2325                                 if (write(fd, null_aligned, togo) < 0)
2326                                         break;
2327                         }
2328                 }
2329                 if (i <= n_config)
2330                         continue;
2331                 d->disk.crc = calc_crc(&d->disk, 512);
2332                 if (write(fd, &d->disk, 512) < 0)
2333                         continue;
2334
2335                 /* Maybe do the same for secondary */
2336
2337                 lseek64(fd, (size-1)*512, SEEK_SET);
2338                 if (write(fd, &ddf->anchor, 512) < 0)
2339                         continue;
2340                 successes++;
2341         }
2342
2343         if (do_close)
2344                 for (d = ddf->dlist; d; d=d->next) {
2345                         close(d->fd);
2346                         d->fd = -1;
2347                 }
2348
2349         return attempts != successes;
2350 }
2351
2352 static int write_init_super_ddf(struct supertype *st)
2353 {
2354         struct ddf_super *ddf = st->sb;
2355         struct vcl *currentconf = ddf->currentconf;
2356
2357         /* we are done with currentconf reset it to point st at the container */
2358         ddf->currentconf = NULL;
2359
2360         if (st->update_tail) {
2361                 /* queue the virtual_disk and vd_config as metadata updates */
2362                 struct virtual_disk *vd;
2363                 struct vd_config *vc;
2364                 int len;
2365
2366                 if (!currentconf) {
2367                         int len = (sizeof(struct phys_disk) +
2368                                    sizeof(struct phys_disk_entry));
2369
2370                         /* adding a disk to the container. */
2371                         if (!ddf->add_list)
2372                                 return 0;
2373
2374                         append_metadata_update(st, ddf->add_list->mdupdate, len);
2375                         ddf->add_list->mdupdate = NULL;
2376                         return 0;
2377                 }
2378
2379                 /* Newly created VD */
2380
2381                 /* First the virtual disk.  We have a slightly fake header */
2382                 len = sizeof(struct virtual_disk) + sizeof(struct virtual_entry);
2383                 vd = malloc(len);
2384                 *vd = *ddf->virt;
2385                 vd->entries[0] = ddf->virt->entries[currentconf->vcnum];
2386                 vd->populated_vdes = __cpu_to_be16(currentconf->vcnum);
2387                 append_metadata_update(st, vd, len);
2388
2389                 /* Then the vd_config */
2390                 len = ddf->conf_rec_len * 512;
2391                 vc = malloc(len);
2392                 memcpy(vc, &currentconf->conf, len);
2393                 append_metadata_update(st, vc, len);
2394
2395                 /* FIXME I need to close the fds! */
2396                 return 0;
2397         } else 
2398                 return __write_init_super_ddf(st, 1);
2399 }
2400
2401 #endif
2402
2403 static __u64 avail_size_ddf(struct supertype *st, __u64 devsize)
2404 {
2405         /* We must reserve the last 32Meg */
2406         if (devsize <= 32*1024*2)
2407                 return 0;
2408         return devsize - 32*1024*2;
2409 }
2410
2411 #ifndef MDASSEMBLE
2412
2413 static int reserve_space(struct supertype *st, int raiddisks,
2414                          unsigned long long size, int chunk,
2415                          unsigned long long *freesize)
2416 {
2417         /* Find 'raiddisks' spare extents at least 'size' big (but
2418          * only caring about multiples of 'chunk') and remember
2419          * them.
2420          * If the cannot be found, fail.
2421          */
2422         struct dl *dl;
2423         struct ddf_super *ddf = st->sb;
2424         int cnt = 0;
2425
2426         for (dl = ddf->dlist; dl ; dl=dl->next) {
2427                 dl->raiddisk = -1;      
2428                 dl->esize = 0;
2429         }
2430         /* Now find largest extent on each device */
2431         for (dl = ddf->dlist ; dl ; dl=dl->next) {
2432                 struct extent *e = get_extents(ddf, dl);
2433                 unsigned long long pos = 0;
2434                 int i = 0;
2435                 int found = 0;
2436                 unsigned long long minsize = size;
2437
2438                 if (size == 0)
2439                         minsize = chunk;
2440
2441                 if (!e)
2442                         continue;
2443                 do {
2444                         unsigned long long esize;
2445                         esize = e[i].start - pos;
2446                         if (esize >= minsize) {
2447                                 found = 1;
2448                                 minsize = esize;
2449                         }
2450                         pos = e[i].start + e[i].size;
2451                         i++;
2452                 } while (e[i-1].size);
2453                 if (found) {
2454                         cnt++;
2455                         dl->esize = minsize;
2456                 }
2457                 free(e);
2458         }
2459         if (cnt < raiddisks) {
2460                 fprintf(stderr, Name ": not enough devices with space to create array.\n");
2461                 return 0; /* No enough free spaces large enough */
2462         }
2463         if (size == 0) {
2464                 /* choose the largest size of which there are at least 'raiddisk' */
2465                 for (dl = ddf->dlist ; dl ; dl=dl->next) {
2466                         struct dl *dl2;
2467                         if (dl->esize <= size)
2468                                 continue;
2469                         /* This is bigger than 'size', see if there are enough */
2470                         cnt = 0;
2471                         for (dl2 = dl; dl2 ; dl2=dl2->next)
2472                                 if (dl2->esize >= dl->esize)
2473                                         cnt++;
2474                         if (cnt >= raiddisks)
2475                                 size = dl->esize;
2476                 }
2477                 if (chunk) {
2478                         size = size / chunk;
2479                         size *= chunk;
2480                 }
2481                 *freesize = size;
2482                 if (size < 32) {
2483                         fprintf(stderr, Name ": not enough spare devices to create array.\n");
2484                         return 0;
2485                 }
2486         }
2487         /* We have a 'size' of which there are enough spaces.
2488          * We simply do a first-fit */
2489         cnt = 0;
2490         for (dl = ddf->dlist ; dl && cnt < raiddisks ; dl=dl->next) {
2491                 if (dl->esize < size)
2492                         continue;
2493                 
2494                 dl->raiddisk = cnt;
2495                 cnt++;
2496         }
2497         return 1;
2498 }
2499
2500
2501
2502 static int
2503 validate_geometry_ddf_container(struct supertype *st,
2504                                 int level, int layout, int raiddisks,
2505                                 int chunk, unsigned long long size,
2506                                 char *dev, unsigned long long *freesize,
2507                                 int verbose);
2508
2509 static int validate_geometry_ddf_bvd(struct supertype *st,
2510                                      int level, int layout, int raiddisks,
2511                                      int chunk, unsigned long long size,
2512                                      char *dev, unsigned long long *freesize,
2513                                      int verbose);
2514
2515 static int validate_geometry_ddf(struct supertype *st,
2516                                  int level, int layout, int raiddisks,
2517                                  int chunk, unsigned long long size,
2518                                  char *dev, unsigned long long *freesize,
2519                                  int verbose)
2520 {
2521         int fd;
2522         struct mdinfo *sra;
2523         int cfd;
2524
2525         /* ddf potentially supports lots of things, but it depends on
2526          * what devices are offered (and maybe kernel version?)
2527          * If given unused devices, we will make a container.
2528          * If given devices in a container, we will make a BVD.
2529          * If given BVDs, we make an SVD, changing all the GUIDs in the process.
2530          */
2531
2532         if (level == LEVEL_CONTAINER) {
2533                 /* Must be a fresh device to add to a container */
2534                 return validate_geometry_ddf_container(st, level, layout,
2535                                                        raiddisks, chunk,
2536                                                        size, dev, freesize,
2537                                                        verbose);
2538         }
2539
2540         if (!dev) {
2541                 /* Initial sanity check.  Exclude illegal levels. */
2542                 int i;
2543                 for (i=0; ddf_level_num[i].num1 != MAXINT; i++)
2544                         if (ddf_level_num[i].num2 == level)
2545                                 break;
2546                 if (ddf_level_num[i].num1 == MAXINT)
2547                         return 0;
2548                 /* Should check layout? etc */
2549
2550                 if (st->sb && freesize) {
2551                         /* --create was given a container to create in.
2552                          * So we need to check that there are enough
2553                          * free spaces and return the amount of space.
2554                          * We may as well remember which drives were
2555                          * chosen so that add_to_super/getinfo_super
2556                          * can return them.
2557                          */
2558                         return reserve_space(st, raiddisks, size, chunk, freesize);
2559                 }
2560                 return 1;
2561         }
2562
2563         if (st->sb) {
2564                 /* A container has already been opened, so we are
2565                  * creating in there.  Maybe a BVD, maybe an SVD.
2566                  * Should make a distinction one day.
2567                  */
2568                 return validate_geometry_ddf_bvd(st, level, layout, raiddisks,
2569                                                  chunk, size, dev, freesize,
2570                                                  verbose);
2571         }
2572         /* This is the first device for the array.
2573          * If it is a container, we read it in and do automagic allocations,
2574          * no other devices should be given.
2575          * Otherwise it must be a member device of a container, and we
2576          * do manual allocation.
2577          * Later we should check for a BVD and make an SVD.
2578          */
2579         fd = open(dev, O_RDONLY|O_EXCL, 0);
2580         if (fd >= 0) {
2581                 sra = sysfs_read(fd, 0, GET_VERSION);
2582                 close(fd);
2583                 if (sra && sra->array.major_version == -1 &&
2584                     strcmp(sra->text_version, "ddf") == 0) {
2585
2586                         /* load super */
2587                         /* find space for 'n' devices. */
2588                         /* remember the devices */
2589                         /* Somehow return the fact that we have enough */
2590                 }
2591
2592                 if (verbose)
2593                         fprintf(stderr,
2594                                 Name ": ddf: Cannot create this array "
2595                                 "on device %s\n",
2596                                 dev);
2597                 return 0;
2598         }
2599         if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
2600                 if (verbose)
2601                         fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2602                                 dev, strerror(errno));
2603                 return 0;
2604         }
2605         /* Well, it is in use by someone, maybe a 'ddf' container. */
2606         cfd = open_container(fd);
2607         if (cfd < 0) {
2608                 close(fd);
2609                 if (verbose)
2610                         fprintf(stderr, Name ": ddf: Cannot use %s: %s\n",
2611                                 dev, strerror(EBUSY));
2612                 return 0;
2613         }
2614         sra = sysfs_read(cfd, 0, GET_VERSION);
2615         close(fd);
2616         if (sra && sra->array.major_version == -1 &&
2617             strcmp(sra->text_version, "ddf") == 0) {
2618                 /* This is a member of a ddf container.  Load the container
2619                  * and try to create a bvd
2620                  */
2621                 struct ddf_super *ddf;
2622                 if (load_super_ddf_all(st, cfd, (void **)&ddf, NULL, 1) == 0) {
2623                         st->sb = ddf;
2624                         st->container_dev = fd2devnum(cfd);
2625                         close(cfd);
2626                         return validate_geometry_ddf_bvd(st, level, layout,
2627                                                          raiddisks, chunk, size,
2628                                                          dev, freesize,
2629                                                          verbose);
2630                 }
2631                 close(cfd);
2632         } else /* device may belong to a different container */
2633                 return 0;
2634
2635         return 1;
2636 }
2637
2638 static int
2639 validate_geometry_ddf_container(struct supertype *st,
2640                                 int level, int layout, int raiddisks,
2641                                 int chunk, unsigned long long size,
2642                                 char *dev, unsigned long long *freesize,
2643                                 int verbose)
2644 {
2645         int fd;
2646         unsigned long long ldsize;
2647
2648         if (level != LEVEL_CONTAINER)
2649                 return 0;
2650         if (!dev)
2651                 return 1;
2652
2653         fd = open(dev, O_RDONLY|O_EXCL, 0);
2654         if (fd < 0) {
2655                 if (verbose)
2656                         fprintf(stderr, Name ": ddf: Cannot open %s: %s\n",
2657                                 dev, strerror(errno));
2658                 return 0;
2659         }
2660         if (!get_dev_size(fd, dev, &ldsize)) {
2661                 close(fd);
2662                 return 0;
2663         }
2664         close(fd);
2665
2666         *freesize = avail_size_ddf(st, ldsize >> 9);
2667         if (*freesize == 0)
2668                 return 0;
2669
2670         return 1;
2671 }
2672
2673 static int validate_geometry_ddf_bvd(struct supertype *st,
2674                                      int level, int layout, int raiddisks,
2675                                      int chunk, unsigned long long size,
2676                                      char *dev, unsigned long long *freesize,
2677                                      int verbose)
2678 {
2679         struct stat stb;
2680         struct ddf_super *ddf = st->sb;
2681         struct dl *dl;
2682         unsigned long long pos = 0;
2683         unsigned long long maxsize;
2684         struct extent *e;
2685         int i;
2686         /* ddf/bvd supports lots of things, but not containers */
2687         if (level == LEVEL_CONTAINER)
2688                 return 0;
2689         /* We must have the container info already read in. */
2690         if (!ddf)
2691                 return 0;
2692
2693         if (!dev) {
2694                 /* General test:  make sure there is space for
2695                  * 'raiddisks' device extents of size 'size'.
2696                  */
2697                 unsigned long long minsize = size;
2698                 int dcnt = 0;
2699                 if (minsize == 0)
2700                         minsize = 8;
2701                 for (dl = ddf->dlist; dl ; dl = dl->next)
2702                 {
2703                         int found = 0;
2704                         pos = 0;
2705
2706                         i = 0;
2707                         e = get_extents(ddf, dl);
2708                         if (!e) continue;
2709                         do {
2710                                 unsigned long long esize;
2711                                 esize = e[i].start - pos;
2712                                 if (esize >= minsize)
2713                                         found = 1;
2714                                 pos = e[i].start + e[i].size;
2715                                 i++;
2716                         } while (e[i-1].size);
2717                         if (found)
2718                                 dcnt++;
2719                         free(e);
2720                 }
2721                 if (dcnt < raiddisks) {
2722                         if (verbose)
2723                                 fprintf(stderr,
2724                                         Name ": ddf: Not enough devices with "
2725                                         "space for this array (%d < %d)\n",
2726                                         dcnt, raiddisks);
2727                         return 0;
2728                 }
2729                 return 1;
2730         }
2731         /* This device must be a member of the set */
2732         if (stat(dev, &stb) < 0)
2733                 return 0;
2734         if ((S_IFMT & stb.st_mode) != S_IFBLK)
2735                 return 0;
2736         for (dl = ddf->dlist ; dl ; dl = dl->next) {
2737                 if (dl->major == major(stb.st_rdev) &&
2738                     dl->minor == minor(stb.st_rdev))
2739                         break;
2740         }
2741         if (!dl) {
2742                 if (verbose)
2743                         fprintf(stderr, Name ": ddf: %s is not in the "
2744                                 "same DDF set\n",
2745                                 dev);
2746                 return 0;
2747         }
2748         e = get_extents(ddf, dl);
2749         maxsize = 0;
2750         i = 0;
2751         if (e) do {
2752                 unsigned long long esize;
2753                 esize = e[i].start - pos;
2754                 if (esize >= maxsize)
2755                         maxsize = esize;
2756                 pos = e[i].start + e[i].size;
2757                 i++;
2758         } while (e[i-1].size);
2759         *freesize = maxsize;
2760         // FIXME here I am
2761
2762         return 1;
2763 }
2764
2765 static int load_super_ddf_all(struct supertype *st, int fd,
2766                               void **sbp, char *devname, int keep_fd)
2767 {
2768         struct mdinfo *sra;
2769         struct ddf_super *super;
2770         struct mdinfo *sd, *best = NULL;
2771         int bestseq = 0;
2772         int seq;
2773         char nm[20];
2774         int dfd;
2775         int devnum = fd2devnum(fd);
2776         enum sysfs_read_flags flags;
2777
2778         flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2779         if (mdmon_running(devnum))
2780                 flags |= SKIP_GONE_DEVS;
2781
2782         sra = sysfs_read(fd, 0, flags);
2783         if (!sra)
2784                 return 1;
2785         if (sra->array.major_version != -1 ||
2786             sra->array.minor_version != -2 ||
2787             strcmp(sra->text_version, "ddf") != 0)
2788                 return 1;
2789
2790         if (posix_memalign((void**)&super, 512, sizeof(*super)) != 0)
2791                 return 1;
2792         memset(super, 0, sizeof(*super));
2793
2794         /* first, try each device, and choose the best ddf */
2795         for (sd = sra->devs ; sd ; sd = sd->next) {
2796                 int rv;
2797                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2798                 dfd = dev_open(nm, O_RDONLY);
2799                 if (dfd < 0)
2800                         return 2;
2801                 rv = load_ddf_headers(dfd, super, NULL);
2802                 close(dfd);
2803                 if (rv == 0) {
2804                         seq = __be32_to_cpu(super->active->seq);
2805                         if (super->active->openflag)
2806                                 seq--;
2807                         if (!best || seq > bestseq) {
2808                                 bestseq = seq;
2809                                 best = sd;
2810                         }
2811                 }
2812         }
2813         if (!best)
2814                 return 1;
2815         /* OK, load this ddf */
2816         sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2817         dfd = dev_open(nm, O_RDONLY);
2818         if (dfd < 0)
2819                 return 1;
2820         load_ddf_headers(dfd, super, NULL);
2821         load_ddf_global(dfd, super, NULL);
2822         close(dfd);
2823         /* Now we need the device-local bits */
2824         for (sd = sra->devs ; sd ; sd = sd->next) {
2825                 int rv;
2826
2827                 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2828                 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2829                 if (dfd < 0)
2830                         return 2;
2831                 rv = load_ddf_headers(dfd, super, NULL);
2832                 if (rv == 0)
2833                         rv = load_ddf_local(dfd, super, NULL, keep_fd);
2834                 if (!keep_fd) close(dfd);
2835                 if (rv)
2836                         return 1;
2837         }
2838         if (st->subarray[0]) {
2839                 struct vcl *v;
2840
2841                 for (v = super->conflist; v; v = v->next)
2842                         if (v->vcnum == atoi(st->subarray))
2843                                 super->currentconf = v;
2844                 if (!super->currentconf)
2845                         return 1;
2846         }
2847         *sbp = super;
2848         if (st->ss == NULL) {
2849                 st->ss = &super_ddf;
2850                 st->minor_version = 0;
2851                 st->max_devs = 512;
2852                 st->container_dev = fd2devnum(fd);
2853         }
2854         st->loaded_container = 1;
2855         return 0;
2856 }
2857 #endif /* MDASSEMBLE */
2858
2859 static struct mdinfo *container_content_ddf(struct supertype *st)
2860 {
2861         /* Given a container loaded by load_super_ddf_all,
2862          * extract information about all the arrays into
2863          * an mdinfo tree.
2864          *
2865          * For each vcl in conflist: create an mdinfo, fill it in,
2866          *  then look for matching devices (phys_refnum) in dlist
2867          *  and create appropriate device mdinfo.
2868          */
2869         struct ddf_super *ddf = st->sb;
2870         struct mdinfo *rest = NULL;
2871         struct vcl *vc;
2872
2873         for (vc = ddf->conflist ; vc ; vc=vc->next)
2874         {
2875                 int i;
2876                 int j;
2877                 struct mdinfo *this;
2878                 this = malloc(sizeof(*this));
2879                 memset(this, 0, sizeof(*this));
2880                 this->next = rest;
2881                 rest = this;
2882
2883                 this->array.level = map_num1(ddf_level_num, vc->conf.prl);
2884                 this->array.raid_disks =
2885                         __be16_to_cpu(vc->conf.prim_elmnt_count);
2886                 this->array.layout = rlq_to_layout(vc->conf.rlq, vc->conf.prl,
2887                                                    this->array.raid_disks);
2888                 this->array.md_minor      = -1;
2889                 this->array.major_version = -1;
2890                 this->array.minor_version = -2;
2891                 this->array.ctime         = DECADE +
2892                         __be32_to_cpu(*(__u32*)(vc->conf.guid+16));
2893                 this->array.utime         = DECADE +
2894                         __be32_to_cpu(vc->conf.timestamp);
2895                 this->array.chunk_size    = 512 << vc->conf.chunk_shift;
2896
2897                 i = vc->vcnum;
2898                 if ((ddf->virt->entries[i].state & DDF_state_inconsistent) ||
2899                     (ddf->virt->entries[i].init_state & DDF_initstate_mask) !=
2900                     DDF_init_full) {
2901                         this->array.state = 0;
2902                         this->resync_start = 0;
2903                 } else {
2904                         this->array.state = 1;
2905                         this->resync_start = ~0ULL;
2906                 }
2907                 memcpy(this->name, ddf->virt->entries[i].name, 16);
2908                 this->name[16]=0;
2909                 for(j=0; j<16; j++)
2910                         if (this->name[j] == ' ')
2911                                 this->name[j] = 0;
2912
2913                 memset(this->uuid, 0, sizeof(this->uuid));
2914                 this->component_size = __be64_to_cpu(vc->conf.blocks);
2915                 this->array.size = this->component_size / 2;
2916                 this->container_member = i;
2917
2918                 ddf->currentconf = vc;
2919                 uuid_from_super_ddf(st, this->uuid);
2920                 ddf->currentconf = NULL;
2921
2922                 sprintf(this->text_version, "/%s/%d",
2923                         devnum2devname(st->container_dev),
2924                         this->container_member);
2925
2926                 for (i=0 ; i < ddf->mppe ; i++) {
2927                         struct mdinfo *dev;
2928                         struct dl *d;
2929
2930                         if (vc->conf.phys_refnum[i] == 0xFFFFFFFF)
2931                                 continue;
2932
2933                         this->array.working_disks++;
2934
2935                         for (d = ddf->dlist; d ; d=d->next)
2936                                 if (d->disk.refnum == vc->conf.phys_refnum[i])
2937                                         break;
2938                         if (d == NULL)
2939                                 /* Haven't found that one yet, maybe there are others */
2940                                 continue;
2941
2942                         dev = malloc(sizeof(*dev));
2943                         memset(dev, 0, sizeof(*dev));
2944                         dev->next = this->devs;
2945                         this->devs = dev;
2946
2947                         dev->disk.number = __be32_to_cpu(d->disk.refnum);
2948                         dev->disk.major = d->major;
2949                         dev->disk.minor = d->minor;
2950                         dev->disk.raid_disk = i;
2951                         dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
2952
2953                         dev->events = __be32_to_cpu(ddf->primary.seq);
2954                         dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
2955                         dev->component_size = __be64_to_cpu(vc->conf.blocks);
2956                         if (d->devname)
2957                                 strcpy(dev->name, d->devname);
2958                 }
2959         }
2960         return rest;
2961 }
2962
2963 static int store_zero_ddf(struct supertype *st, int fd)
2964 {
2965         unsigned long long dsize;
2966         void *buf;
2967         int rc;
2968
2969         if (!get_dev_size(fd, NULL, &dsize))
2970                 return 1;
2971
2972         if (posix_memalign(&buf, 512, 512) != 0)
2973                 return 1;
2974         memset(buf, 0, 512);
2975
2976         lseek64(fd, dsize-512, 0);
2977         rc = write(fd, buf, 512);
2978         free(buf);
2979         if (rc < 0)
2980                 return 1;
2981         return 0;
2982 }
2983
2984 static int compare_super_ddf(struct supertype *st, struct supertype *tst)
2985 {
2986         /*
2987          * return:
2988          *  0 same, or first was empty, and second was copied
2989          *  1 second had wrong number
2990          *  2 wrong uuid
2991          *  3 wrong other info
2992          */
2993         struct ddf_super *first = st->sb;
2994         struct ddf_super *second = tst->sb;
2995
2996         if (!first) {
2997                 st->sb = tst->sb;
2998                 tst->sb = NULL;
2999                 return 0;
3000         }
3001
3002         if (memcmp(first->anchor.guid, second->anchor.guid, DDF_GUID_LEN) != 0)
3003                 return 2;
3004
3005         /* FIXME should I look at anything else? */
3006         return 0;
3007 }
3008
3009 #ifndef MDASSEMBLE
3010 /*
3011  * A new array 'a' has been started which claims to be instance 'inst'
3012  * within container 'c'.
3013  * We need to confirm that the array matches the metadata in 'c' so
3014  * that we don't corrupt any metadata.
3015  */
3016 static int ddf_open_new(struct supertype *c, struct active_array *a, char *inst)
3017 {
3018         dprintf("ddf: open_new %s\n", inst);
3019         a->info.container_member = atoi(inst);
3020         return 0;
3021 }
3022
3023 /*
3024  * The array 'a' is to be marked clean in the metadata.
3025  * If '->resync_start' is not ~(unsigned long long)0, then the array is only
3026  * clean up to the point (in sectors).  If that cannot be recorded in the
3027  * metadata, then leave it as dirty.
3028  *
3029  * For DDF, we need to clear the DDF_state_inconsistent bit in the
3030  * !global! virtual_disk.virtual_entry structure.
3031  */
3032 static int ddf_set_array_state(struct active_array *a, int consistent)
3033 {
3034         struct ddf_super *ddf = a->container->sb;
3035         int inst = a->info.container_member;
3036         int old = ddf->virt->entries[inst].state;
3037         if (consistent == 2) {
3038                 /* Should check if a recovery should be started FIXME */
3039                 consistent = 1;
3040                 if (!is_resync_complete(a))
3041                         consistent = 0;
3042         }
3043         if (consistent)
3044                 ddf->virt->entries[inst].state &= ~DDF_state_inconsistent;
3045         else
3046                 ddf->virt->entries[inst].state |= DDF_state_inconsistent;
3047         if (old != ddf->virt->entries[inst].state)
3048                 ddf->updates_pending = 1;
3049
3050         old = ddf->virt->entries[inst].init_state;
3051         ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
3052         if (is_resync_complete(a))
3053                 ddf->virt->entries[inst].init_state |= DDF_init_full;
3054         else if (a->resync_start == 0)
3055                 ddf->virt->entries[inst].init_state |= DDF_init_not;
3056         else
3057                 ddf->virt->entries[inst].init_state |= DDF_init_quick;
3058         if (old != ddf->virt->entries[inst].init_state)
3059                 ddf->updates_pending = 1;
3060
3061         dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
3062                 a->resync_start);
3063         return consistent;
3064 }
3065
3066 /*
3067  * The state of each disk is stored in the global phys_disk structure
3068  * in phys_disk.entries[n].state.
3069  * This makes various combinations awkward.
3070  * - When a device fails in any array, it must be failed in all arrays
3071  *   that include a part of this device.
3072  * - When a component is rebuilding, we cannot include it officially in the
3073  *   array unless this is the only array that uses the device.
3074  *
3075  * So: when transitioning:
3076  *   Online -> failed,  just set failed flag.  monitor will propagate
3077  *   spare -> online,   the device might need to be added to the array.
3078  *   spare -> failed,   just set failed.  Don't worry if in array or not.
3079  */
3080 static void ddf_set_disk(struct active_array *a, int n, int state)
3081 {
3082         struct ddf_super *ddf = a->container->sb;
3083         int inst = a->info.container_member;
3084         struct vd_config *vc = find_vdcr(ddf, inst);
3085         int pd = find_phys(ddf, vc->phys_refnum[n]);
3086         int i, st, working;
3087
3088         if (vc == NULL) {
3089                 dprintf("ddf: cannot find instance %d!!\n", inst);
3090                 return;
3091         }
3092         if (pd < 0) {
3093                 /* disk doesn't currently exist. If it is now in_sync,
3094                  * insert it. */
3095                 if ((state & DS_INSYNC) && ! (state & DS_FAULTY)) {
3096                         /* Find dev 'n' in a->info->devs, determine the
3097                          * ddf refnum, and set vc->phys_refnum and update
3098                          * phys->entries[]
3099                          */
3100                         /* FIXME */
3101                 }
3102         } else {
3103                 int old = ddf->phys->entries[pd].state;
3104                 if (state & DS_FAULTY)
3105                         ddf->phys->entries[pd].state  |= __cpu_to_be16(DDF_Failed);
3106                 if (state & DS_INSYNC) {
3107                         ddf->phys->entries[pd].state  |= __cpu_to_be16(DDF_Online);
3108                         ddf->phys->entries[pd].state  &= __cpu_to_be16(~DDF_Rebuilding);
3109                 }
3110                 if (old != ddf->phys->entries[pd].state)
3111                         ddf->updates_pending = 1;
3112         }
3113
3114         dprintf("ddf: set_disk %d to %x\n", n, state);
3115
3116         /* Now we need to check the state of the array and update
3117          * virtual_disk.entries[n].state.
3118          * It needs to be one of "optimal", "degraded", "failed".
3119          * I don't understand 'deleted' or 'missing'.
3120          */
3121         working = 0;
3122         for (i=0; i < a->info.array.raid_disks; i++) {
3123                 pd = find_phys(ddf, vc->phys_refnum[i]);
3124                 if (pd < 0)
3125                         continue;
3126                 st = __be16_to_cpu(ddf->phys->entries[pd].state);
3127                 if ((st & (DDF_Online|DDF_Failed|DDF_Rebuilding))
3128                     == DDF_Online)
3129                         working++;
3130         }
3131         state = DDF_state_degraded;
3132         if (working == a->info.array.raid_disks)
3133                 state = DDF_state_optimal;
3134         else switch(vc->prl) {
3135         case DDF_RAID0:
3136         case DDF_CONCAT:
3137         case DDF_JBOD:
3138                 state = DDF_state_failed;
3139                 break;
3140         case DDF_RAID1:
3141                 if (working == 0)
3142                         state = DDF_state_failed;
3143                 break;
3144         case DDF_RAID4:
3145         case DDF_RAID5:
3146                 if (working < a->info.array.raid_disks-1)
3147                         state = DDF_state_failed;
3148                 break;
3149         case DDF_RAID6:
3150                 if (working < a->info.array.raid_disks-2)
3151                         state = DDF_state_failed;
3152                 else if (working == a->info.array.raid_disks-1)
3153                         state = DDF_state_part_optimal;
3154                 break;
3155         }
3156
3157         if (ddf->virt->entries[inst].state !=
3158             ((ddf->virt->entries[inst].state & ~DDF_state_mask)
3159              | state)) {
3160
3161                 ddf->virt->entries[inst].state =
3162                         (ddf->virt->entries[inst].state & ~DDF_state_mask)
3163                         | state;
3164                 ddf->updates_pending = 1;
3165         }
3166
3167 }
3168
3169 static void ddf_sync_metadata(struct supertype *st)
3170 {
3171
3172         /*
3173          * Write all data to all devices.
3174          * Later, we might be able to track whether only local changes
3175          * have been made, or whether any global data has been changed,
3176          * but ddf is sufficiently weird that it probably always
3177          * changes global data ....
3178          */
3179         struct ddf_super *ddf = st->sb;
3180         if (!ddf->updates_pending)
3181                 return;
3182         ddf->updates_pending = 0;
3183         __write_init_super_ddf(st, 0);
3184         dprintf("ddf: sync_metadata\n");
3185 }
3186
3187 static void ddf_process_update(struct supertype *st,
3188                                struct metadata_update *update)
3189 {
3190         /* Apply this update to the metadata.
3191          * The first 4 bytes are a DDF_*_MAGIC which guides
3192          * our actions.
3193          * Possible update are:
3194          *  DDF_PHYS_RECORDS_MAGIC
3195          *    Add a new physical device.  Changes to this record
3196          *    only happen implicitly.
3197          *    used_pdes is the device number.
3198          *  DDF_VIRT_RECORDS_MAGIC
3199          *    Add a new VD.  Possibly also change the 'access' bits.
3200          *    populated_vdes is the entry number.
3201          *  DDF_VD_CONF_MAGIC
3202          *    New or updated VD.  the VIRT_RECORD must already
3203          *    exist.  For an update, phys_refnum and lba_offset
3204          *    (at least) are updated, and the VD_CONF must
3205          *    be written to precisely those devices listed with
3206          *    a phys_refnum.
3207          *  DDF_SPARE_ASSIGN_MAGIC
3208          *    replacement Spare Assignment Record... but for which device?
3209          *
3210          * So, e.g.:
3211          *  - to create a new array, we send a VIRT_RECORD and
3212          *    a VD_CONF.  Then assemble and start the array.
3213          *  - to activate a spare we send a VD_CONF to add the phys_refnum
3214          *    and offset.  This will also mark the spare as active with
3215          *    a spare-assignment record.
3216          */
3217         struct ddf_super *ddf = st->sb;
3218         __u32 *magic = (__u32*)update->buf;
3219         struct phys_disk *pd;
3220         struct virtual_disk *vd;
3221         struct vd_config *vc;
3222         struct vcl *vcl;
3223         struct dl *dl;
3224         int mppe;
3225         int ent;
3226
3227         dprintf("Process update %x\n", *magic);
3228
3229         switch (*magic) {
3230         case DDF_PHYS_RECORDS_MAGIC:
3231
3232                 if (update->len != (sizeof(struct phys_disk) +
3233                                     sizeof(struct phys_disk_entry)))
3234                         return;
3235                 pd = (struct phys_disk*)update->buf;
3236
3237                 ent = __be16_to_cpu(pd->used_pdes);
3238                 if (ent >= __be16_to_cpu(ddf->phys->max_pdes))
3239                         return;
3240                 if (!all_ff(ddf->phys->entries[ent].guid))
3241                         return;
3242                 ddf->phys->entries[ent] = pd->entries[0];
3243                 ddf->phys->used_pdes = __cpu_to_be16(1 +
3244                                            __be16_to_cpu(ddf->phys->used_pdes));
3245                 ddf->updates_pending = 1;
3246                 if (ddf->add_list) {
3247                         struct active_array *a;
3248                         struct dl *al = ddf->add_list;
3249                         ddf->add_list = al->next;
3250
3251                         al->next = ddf->dlist;
3252                         ddf->dlist = al;
3253
3254                         /* As a device has been added, we should check
3255                          * for any degraded devices that might make
3256                          * use of this spare */
3257                         for (a = st->arrays ; a; a=a->next)
3258                                 a->check_degraded = 1;
3259                 }
3260                 break;
3261
3262         case DDF_VIRT_RECORDS_MAGIC:
3263
3264                 if (update->len != (sizeof(struct virtual_disk) +
3265                                     sizeof(struct virtual_entry)))
3266                         return;
3267                 vd = (struct virtual_disk*)update->buf;
3268
3269                 ent = __be16_to_cpu(vd->populated_vdes);
3270                 if (ent >= __be16_to_cpu(ddf->virt->max_vdes))
3271                         return;
3272                 if (!all_ff(ddf->virt->entries[ent].guid))
3273                         return;
3274                 ddf->virt->entries[ent] = vd->entries[0];
3275                 ddf->virt->populated_vdes = __cpu_to_be16(1 +
3276                               __be16_to_cpu(ddf->virt->populated_vdes));
3277                 ddf->updates_pending = 1;
3278                 break;
3279
3280         case DDF_VD_CONF_MAGIC:
3281                 dprintf("len %d %d\n", update->len, ddf->conf_rec_len);
3282
3283                 mppe = __be16_to_cpu(ddf->anchor.max_primary_element_entries);
3284                 if (update->len != ddf->conf_rec_len * 512)
3285                         return;
3286                 vc = (struct vd_config*)update->buf;
3287                 for (vcl = ddf->conflist; vcl ; vcl = vcl->next)
3288                         if (memcmp(vcl->conf.guid, vc->guid, DDF_GUID_LEN) == 0)
3289                                 break;
3290                 dprintf("vcl = %p\n", vcl);
3291                 if (vcl) {
3292                         /* An update, just copy the phys_refnum and lba_offset
3293                          * fields
3294                          */
3295                         memcpy(vcl->conf.phys_refnum, vc->phys_refnum,
3296                                mppe * (sizeof(__u32) + sizeof(__u64)));
3297                 } else {
3298                         /* A new VD_CONF */
3299                         if (!u